/* ===========================================================================
   Vazhai Chit — module screens (B1)
   ---------------------------------------------------------------------------
   The shared vocabulary every module list, toolbar, table and pager is built
   from. Loaded by `layouts/app.php` after `theme.css`, which owns the tokens;
   NOTHING here introduces a colour. Every value is a `--vz-*` custom property,
   so the dark palette and any future white-label swap apply for free.

   ── WHY THIS IS AN EXTERNAL FILE ─────────────────────────────────────────
   `PHASE_B §B-JS` puts JavaScript in external files so a future
   Content-Security-Policy needs no `unsafe-inline`. The same argument applies
   to CSS, byte for byte: an inline <style> block needs `style-src
   'unsafe-inline'`, which is the concession that usually makes a CSP
   worthless. Phase A's layouts carry inline <style>; from here new component
   CSS lands in a file, and `asset()` cache-busts it by mtime exactly as it
   does theme.css.

   ── NAMING: `.vz-action` IS NOW THE ONLY BUTTON SYSTEM (✅ closed in E3) ───
   B1 recorded this as debt in its own words: *"`.vz-btn` already exists inside
   `layouts/auth.php` as a FULL-WIDTH form submit — the right shape for a
   centred login card, the wrong one for a toolbar … when the auth layout's CSS
   moves out of its <style> block the two should be reconciled into one button
   system."*

   E3 is that moment. The auth layout loads this file, `.vz-btn` is gone, and
   the full-width shape it existed for is `.vz-action--block`.

   ⚠ AND THE DEBT HAD ALREADY COST SOMETHING. `chits/auction-conduct.php` — the
   AUCTION CLOSE, a money screen — rendered `class="vz-btn vz-btn--primary"`
   inside the APP layout, where neither name has ever had a rule. That button
   drew as an unstyled browser default from C1b until E3. Nobody reported it,
   because an unstyled button is still a button: it clicks, it submits, it
   passes every overflow, tap-target and contrast check. Found by the modifier
   scan in `StylesheetStaticTest`, not by looking at the screen.
   =========================================================================== */

/* --- Page header ---------------------------------------------------------- */

.vz-page {
  display: flex;
  flex-wrap: wrap;
  align-items: flex-end;
  justify-content: space-between;
  gap: var(--vz-space-3);
  margin-bottom: var(--vz-space-5);
}

/* 🐛 ⚠ THE HEADING PAIR OWNS ITS OWN ROOM NOW, AND E6+ IS WHY IT HAS TO (E6b). Item 12
   found the auctions diary sitting 4px above its table and fixed it by putting that
   heading inside `.vz-page`, whose `margin-bottom` is where the 20px comes from — then
   claimed the result was *"checked on every screen"*. **The OVERFLOW was checked on every
   screen; the GAP was checked on the one that was reported.** The owner found the rest
   the same day: **25 views render a page heading with no `.vz-page` around it**, every one
   of them 4px above whatever follows.

   ✅ SO THE FIX MOVES ONTO THE COMPONENTS THEMSELVES, WHERE IT CANNOT BE FORGOTTEN. A
   title and a subtitle each carry the space that belongs beneath them, and `.vz-page`
   takes it back off when it is supplying that space itself. **25 views change with no
   markup edited at all**, which also means no document layout is touched by hand.

   ⚠ THE `:has()` IS AN ENHANCEMENT, exactly as `.vz-panel-title`'s is. Without it a title
   followed by a subtitle gets 20px between them instead of 4 — looser, never broken. */
.vz-page-title {
  margin: 0 0 var(--vz-space-5);
  font-size: var(--vz-fs-h1);
  line-height: var(--vz-lh-tight);
  /* Large text (24px) clears the 3.0 bar on brand green in both themes; small
     text does not, which is why nothing else on this page is green. */
  color: var(--vz-primary-strong);
}

/* The pair: 4px between them, and the room below belongs to the second one. */
.vz-page-title:has(+ .vz-page-sub) { margin-bottom: 0; }

.vz-page-sub {
  margin: var(--vz-space-1) 0 var(--vz-space-5);
  color: var(--vz-subtle);
  font-size: var(--vz-fs-sm);
}

/* ⚠ AND INSIDE THE PAGE HEADER THE SPACE IS `.vz-page`'s, so the pair gives it back —
   otherwise every list gains 20px of header it did not have yesterday. These two rules
   come after the `:has()` deliberately: same specificity, and this is the one that has to
   win inside `.vz-page`. */
.vz-page .vz-page-title,
.vz-page .vz-page-sub { margin-bottom: 0; }

.vz-page-actions {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2);
}

/* 🔴 **THE ACTION ROW HAD NO GAP BENEATH IT, ON 44 OF THE 73 SCREENS THAT DRAW ONE.**
   Owner, 2026-10-13, on `/app/compliance`: *"generate the calendar and add a filing button has
   not space below"*. ▶ It is not that screen's bug: `.vz-page-actions` carries no margin at
   all, so whether a gap appears depends entirely on what happens to come next.

   ⚠ **MEASURED, BECAUSE "WHAT COMES NEXT" IS A POPULATION AND NOT A GUESS.** Every view
   rendering the row was scanned and each follower's own `margin-top` read out of this file:

   | flush at 0px — patched here | | already spaced — LEFT ALONE | |
   |---|---|---|---|
   | `.vz-notice`      | 20 | `.vz-panel`        | 16 |
   | `.vz-callout`     | 19 | `.vz-sections`     | 16 |
   | `.vz-toolbar`     |  2 | `.vz-page-sub`     |  4 |
   | `.vz-empty`       |  2 | `.vz-form-note`    | 16 |
   | `.vz-panel-title` |  1 | `.vz-form-actions` | 20 |
   | `.vz-result-bar`  |  1 | | |
   | `.vz-steps`       |  1 | | |

   ⛔ **SO NOT A BLANKET `margin-bottom` ON THE ROW**, which is the obvious fix and the wrong
   one: it would ADD a second gap to the five on the right, which already carry considered
   values of their own. That is the same refusal the `.vz-kv` note two thousand lines down
   reaches, for the same reason, and `a-declaration-is-not-a-rule`.

   ⚠ **16px IS INHERITED, NOT CHOSEN.** `.vz-panel` and `.vz-sections` already receive
   `--vz-space-4` in this exact position, so the patched pairs are being brought up to what the
   unpatched ones have had all along — rather than to a number somebody liked.

   ⚠ **THE COMPLIANCE SCREEN NEEDS TWO OF THESE, AND THE SCAN COULD ONLY SEE ONE.** Its export
   bar arrives through `$this->render()`, so no `class="vz-toolbar"` appears in that template
   and a text scan cannot find it — `scanning-php-templates`. The `.vz-toolbar` pair is in this
   list because the partial was read, not because the scan reported it. */
.vz-page-actions + .vz-notice,
.vz-page-actions + .vz-callout,
.vz-page-actions + .vz-toolbar,
.vz-page-actions + .vz-empty,
.vz-page-actions + .vz-panel-title,
.vz-page-actions + .vz-result-bar,
.vz-page-actions + .vz-steps {
  margin-top: var(--vz-space-4);
}

/* --- Actions -------------------------------------------------------------- */

.vz-action {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  gap: var(--vz-space-2);
  min-height: var(--vz-tap-min);
  padding: 0 var(--vz-space-4);
  box-sizing: border-box;
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  color: var(--vz-ink);
  font-family: var(--vz-font-sans);
  font-size: var(--vz-fs-sm);
  font-weight: var(--vz-fw-medium);
  text-decoration: none;
  cursor: pointer;
}

.vz-action:hover { background: var(--vz-surface-2); }

.vz-action:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* ⚠ FINANCE'S `.vz-btn--primary` IN EVERY RESPECT BUT THE FILL'S LIGHTNESS (E15). Same
   structure, same weight, same radius, same icon, same darken-on-hover, same WHITE label —
   and a fill two steps down Finance's own hue so that label clears the bar (4.96 / 6.48
   against 2.59 / 4.34). `theme.css` carries the measurement and the four candidates it was
   chosen from; the short version is that white text is what the owner wanted kept, and a
   white label is what decides how dark the fill has to be. */
.vz-action--primary {
  border-color: transparent;
  background: var(--vz-primary-button);
  color: var(--vz-primary-contrast);
}

.vz-action--primary:hover {
  background: var(--vz-primary-hover);
  color: var(--vz-primary-contrast);
}

/* 🐛 ⚠ `.vz-action--small` WAS DELETED HERE, AND THE ROW THAT DELETED IT IS THE ROW THAT
   MADE IT DEAD (E6+ item 9). E3 wrote the rule because `auctions/index.php` had rendered
   the class since C1b with no rule anywhere — a "small" button drawing at full size,
   found by the modifier scan rather than by looking. Item 9 replaced that one call site
   with `.vz-action--row`, leaving the only user of the modifier gone.

   ⚠ IT IS DELETED RATHER THAN KEPT FOR LATER, ON THIS ROW'S OWN ARGUMENT. Item 3 of
   `§E29` is three declarations that could never fire and were believed to work for
   thirteen rows; a modifier no view renders is the same thing one step earlier.
   `testEveryModifierRenderedByAViewHasACssRule` guards the other direction and would not
   have said a word. If a compact action is wanted again, the reasoning worth keeping is
   the one that rule carried: **smaller means less padding and smaller text, never a
   shorter button** — `--vz-tap-min` is a floor, not a default. */
/* Full width — the single decisive action on a narrow, one-column surface (E3).
   The auth screens are the case: a sign-in button that shrink-wraps its label in a
   360px card reads as secondary, which is the opposite of what it is. Added when the
   auth layout stopped carrying its own `.vz-btn` and joined this vocabulary. */
.vz-action--block { width: 100%; }

/* A destination that is not the main one on the screen: no box, just a link
   that still meets the 44px tap target. */
.vz-action--quiet {
  border-color: transparent;
  background: none;
  padding: 0 var(--vz-space-2);
  text-decoration: underline;
  color: var(--vz-ink);
  font-weight: var(--vz-fw-regular);
}

.vz-action--quiet:hover { background: none; color: var(--vz-primary-strong); }

/* ===========================================================================
   A table row's action — a glyph on a screen, a word on a phone (E6+ item 9)
   ---------------------------------------------------------------------------
   ⚠ 44 × 44, WHICH IS THE WHOLE POINT OF `width` HERE. `.vz-action` already gives the
   height; without a width an icon-only button shrink-wraps a 20px glyph inside 32px of
   padding and lands under the floor. `padding: 0` and an explicit square is the only
   arrangement that is 44 in both directions whatever the glyph.

   ⚠ THE LABEL IS `display: none` ABOVE 44rem, NOT VISUALLY HIDDEN. Hidden text stays in
   the accessibility tree and would be announced ALONGSIDE the `aria-label` that carries
   the same words — the duplicated-accessible-name defect this repo has now hit three
   times. Removed from the tree, `aria-label` is the single name; below 44rem the label is
   real text and `aria-label` repeats it exactly, which is WCAG 2.5.3 satisfied rather
   than merely not violated.

   ⚠ AND `text-decoration: none`, BECAUSE `--quiet` UNDERLINES. An underline under a
   glyph is a rule with no word on it. `partials/row-action.php` carries the rest of the
   argument, including which actions are deliberately NOT glyphs.
   =========================================================================== */
.vz-action--row {
  width: var(--vz-tap-min);
  padding: 0;
  color: var(--vz-muted);
  text-decoration: none;
}

.vz-action--row:hover {
  background: var(--vz-surface-2);
  color: var(--vz-ink);
}

.vz-action--row .vz-action-label { display: none; }

/* Several glyphs in one cell need to breathe; `.vz-c-actions` is `nowrap` by default so
   they sit on one line and this is the gap between them. */
.vz-c-actions .vz-action--row + .vz-action--row { margin-left: var(--vz-space-1); }

@media (max-width: 64rem) {
  /* ⚠ `title` DOES NOT EXIST ON TOUCH. There is no hover on a phone, so an icon-only
     control there is an unlabelled one — and this is the surface a collector uses at
     somebody's door. The table is already a stack of cards at this width (`§E21`) and
     the card has the room, so the word comes back and the button goes back to being an
     ordinary `--quiet` action. */
  .vz-action--row {
    width: auto;
    padding: 0 var(--vz-space-2);
    color: var(--vz-ink);
    text-decoration: underline;
  }

  /* `block` rather than `inline`, because the button is a flex container and a flex
     item is blockified whatever this says — writing `inline` would be a declaration
     the browser silently rewrites, which is one more thing that does not mean what it
     appears to. What matters is that it is not `none`: `display: none` removes the
     node from the accessibility tree, and above this breakpoint that is deliberate. */
  .vz-action--row .vz-action-label { display: block; }
}

/* ===========================================================================
   Destructive actions (E17, owner-approved from rendered examples: option 3)
   ---------------------------------------------------------------------------
   ⚠ THE PRODUCT HAD NO DANGER STYLING AT ALL, ON A MONEY SYSTEM. Reversing a
   settlement — money back out, reversing entries into an append-only ledger —
   rendered as `class="vz-action"`, identical to *Cancel* beside it. Reversing a
   whole collection batch, and voiding a payment somebody holds a receipt for,
   rendered as `.vz-action--quiet`: **the least prominent control on the screen
   was the most consequential one**, sitting in the same table cell as the benign
   *Clear* button and looking exactly like it.

   `--vz-danger` existed as a token from S5 and painted the required asterisk and
   error text. `vazhai-finance` cannot supply the answer here — it has no danger
   button either — so this is the one component in the design pass that is Chit's
   own rather than ported.

   ── ⚠ WHY TWO VARIANTS, AND WHICH ACT GETS WHICH ──────────────────────────
   The owner's rule is that **the weight escalates as the action gets closer to
   irreversible**. This product has no two-stage trigger-then-confirm flow — every
   destructive act is one form on the page — so the escalation is drawn where the
   product's own consequences differ:

     `--danger`        the act stops something that has NOT happened yet, or
                       removes a record no money depends on. Refuse a payout or a
                       settlement, discard a DRAFT group, delete an unfiled
                       compliance row, remove a contact, a nominee, a file, a seat.

     `--danger-solid`  the act takes back money that HAS ALREADY MOVED, or
                       destroys a receipt somebody is holding. Void or bounce a
                       payment, reverse a collection batch, reverse a settlement.

   ⚠ **REFUSE IS DELIBERATELY *NOT* SOLID**, and that is the whole reason option 2
   was rejected: rendered side by side, a solid red out-competes *Approve & pay*,
   so the most dangerous control becomes the most attention-grabbing thing on a
   screen where the safe action is usually the one you want. Refuse and approve sit
   on the SAME screen in `payout.php` and `settlements/show.php`. A refusal also
   undoes nothing — `PHASE_C §C4e` records that a refusal puts the seat back while a
   reversal leaves it retired — so weighting them alike would say something false.
   =========================================================================== */

/* Outlined: the destructive act is CHOSEN here. Reads as a real warning without
   competing with the primary action beside it. Label 10.35 on the surface, 8.98 on
   its own hover tint; the border is 3.93, which is a non-text floor of 3.0. */
.vz-action--danger {
  border-color: var(--vz-danger);
  background: var(--vz-surface);
  color: var(--vz-danger-text);
}

.vz-action--danger:hover {
  background: var(--vz-danger-tint);
  color: var(--vz-danger-text);
}

/* Solid: money already moved, and this button takes it back. The heaviest control
   in the product, and the only one that is. */
.vz-action--danger-solid {
  border-color: transparent;
  background: var(--vz-danger-button);
  color: var(--vz-primary-contrast);
}

.vz-action--danger-solid:hover {
  background: var(--vz-danger-hover);
  color: var(--vz-primary-contrast);
}

/* ⚠ THE FOCUS RING STAYS `--vz-primary` ON BOTH, AND THAT IS DELIBERATE. A focus
   indicator answers "where am I", not "what does this do", and one that changes
   colour per control is one a keyboard user has to learn twice. `outline-offset: 2px`
   puts it on the page ground rather than on the red fill, so it is legible on both. */

.vz-action[disabled],
.vz-action[aria-disabled='true'] {
  opacity: 0.55;
  cursor: not-allowed;
}

/* ⚠ THE BUTTON THAT HAS ALREADY BEEN PRESSED (E19). The form carries the state
   and this only shows it — see `app.js` for why it is not `disabled`. The cursor
   says "waiting" rather than "refused", which is the difference between this and
   the rule above: the action is happening, not blocked. It inherits that rule's
   dimming through `aria-disabled`, so a busy control reads as inactive too. */
.vz-action.is-busy { cursor: wait; }

/* --- Toolbar: search + filters -------------------------------------------- */

.vz-toolbar {
  display: flex;
  flex-wrap: wrap;
  align-items: flex-end;
  gap: var(--vz-space-3);
  margin-bottom: var(--vz-space-4);
  padding: var(--vz-space-4);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-lg);
  background: var(--vz-surface);
}

.vz-toolbar-field {
  display: flex;
  flex-direction: column;
  gap: var(--vz-space-1);
  /* Grows to fill the bar on a wide screen, wraps to full width on a narrow one
     rather than shrinking to an unusable stub. */
  flex: 1 1 12rem;
  min-width: 0;
}

.vz-toolbar-field--search { flex: 2 1 18rem; }

/* 🔴 A FIELD LEFT ALONE ON THE LAST FLEX LINE STRETCHES TO FILL IT, AND ON A SELECT OF
   TWO-CHARACTER VALUES THAT IS ABSURD (`§7` item 104, owner-approved 2026-09-17).
   Measured, shipped: `/app/audit`'s **To** date box is **841px at 1440**; `/app/members`'
   **KYC** select is **610px at 1024** and **539px at 705**. Nothing overflows and nothing
   is crushed — `§7` 101's family: the box fits, and it is the wrong size for what is in it.

   ⚠ 18rem IS NOT AN INVENTED NUMBER — IT IS THE SEARCH FIELD'S OWN BASIS, one line up. The
   rule reads *"no filter is ever wider than the search box asks to be"*, and it was chosen
   by measurement rather than taste: at 12rem every field is pinned to its basis and widths
   that are fine today move (audit's Module 260 → 192 at 1024); at 24rem a two-character
   select still sits in a 384px box. **18rem moves the stretched fields and nothing else.**

   ⛔ THE SEARCH BOX IS EXEMPT AND MUST STAY EXEMPT. It is the one control on the bar that
   benefits from every pixel — people type names into it — and capping it would take
   `/app/members`' search from 360 to 288 at 1440 for no reason.

   🔴 44.0625rem IS ONE PIXEL ABOVE THE TOOLBAR'S OWN BREAKPOINT, AND IT IS ALSO EXACTLY
   WHERE THE DEFECT STARTS. Two independent reasons for one number, both measured across six
   captured bars: fields begin sharing a line at **469px**, the first stretched lone field
   appears at **705px** (821 on audit), and `@media (max-width: 44rem)` already owns the
   narrow toolbar. ⛔ **Below it the cap must NOT apply**: every field is one per line there
   and filling the line is correct — capping at 375 insets each stacked select to 288 inside
   a 328px bar while the exempt search stays 302, which is a ragged bar rather than a fix.

   ✅ MEASURED SURGICAL, with the rule injected and removed: `/app/members` unchanged at 320,
   375, 480, 704, 1280 and 1440, moving only at 705 (539 → 288) and 1024 (610 → 288);
   `/app/audit` unchanged at everything up to 1024 and moving only at 1280/1440 (841 → 288).
   ⚠ **AND `/app/members` IS THE CONTROL ITEM 104 ASKED FOR** — a screen that row never
   touched HAS to move, which is the opposite of the usual control and the proof the fix
   reaches past the one bar it was reported on. */
@media (min-width: 44.0625rem) {
  .vz-toolbar-field { max-width: 18rem; }
  .vz-toolbar-field--search { max-width: none; }
}

.vz-toolbar-label {
  font-size: var(--vz-fs-xs);
  font-weight: var(--vz-fw-medium);
  letter-spacing: 0.04em;
  text-transform: uppercase;
  color: var(--vz-subtle);
}

.vz-control {
  width: 100%;
  min-height: var(--vz-tap-min);
  padding: 0 var(--vz-space-3);
  box-sizing: border-box;
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  color: var(--vz-ink);
  font-family: var(--vz-font-sans);
  font-size: var(--vz-fs-md);
}

/* 🔴 **THE FILE PICKER IS THE ONE CONTROL ON ANY OF THESE SCREENS THE BROWSER DRAWS**
   (owner, 2026-10-13, from `/app/compliance/{id}`). Measured there: a **960×44** bordered
   box holding a **~187px** native widget — four fifths of it empty — with Chrome's own grey
   *Choose File* chrome sitting inside a page whose every other button is a `.vz-action`.

   ▶ `::file-selector-button` is the only part of that widget CSS can reach, and giving it
   the `.vz-action` look is what makes the control read as part of the product. ⛔ The
   alternative is the usual trick — hide the input, label a button, wire it up in script —
   and this page ships no JavaScript and says so.

   ⚠ **IT IS AN ATTRIBUTE SELECTOR ON `.vz-control`, NOT A NEW MODIFIER, BECAUSE THE DEFECT
   IS PRODUCT-WIDE AND A COMPLIANCE-ONLY CLASS WOULD BE THE DRIFT THIS FILE EXISTS TO STOP.**
   Eight sites render a file input: compliance, `members/file-form`, `support/new`,
   `support/show` through `partials/field`, and `onboarding`, `profile` and `settings` with
   the class written out. Four of them would have kept the browser's button.

   ⚠ THE BUTTON IS SHORTER THAN `--vz-tap-min` AND THAT IS NOT AN OVERSIGHT: it sits INSIDE a
   44px control, and the whole box opens the picker, so the tap target is the row rather than
   the button drawn on it. */
input[type="file"].vz-control {
  padding-inline: var(--vz-space-2);
  font-size: var(--vz-fs-sm);
  color: var(--vz-muted);
  cursor: pointer;
}

input[type="file"].vz-control::file-selector-button {
  min-height: 2rem;
  margin-inline-end: var(--vz-space-3);
  padding: 0 var(--vz-space-3);
  box-sizing: border-box;
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-sm);
  background: var(--vz-surface-2);
  color: var(--vz-ink);
  font-family: var(--vz-font-sans);
  font-size: var(--vz-fs-sm);
  font-weight: var(--vz-fw-medium);
  cursor: pointer;
}

input[type="file"].vz-control::file-selector-button:hover { background: var(--vz-border); }

/* ⚠ FINANCE'S FOCUS RING, MATCHED (E5). The reported difference: chit drew a hard 2px
   outline offset off the control, Finance draws a soft 3px TINTED ring flush to the
   border and transitions into it. Same two signals either way — the border changes
   colour AND a ring appears — so the focus indicator is still two independent cues,
   which is what `outline: none` has to be justified by. */
.vz-control {
  transition: border-color .15s ease, box-shadow .15s ease;
}

.vz-control:focus,
.vz-control:focus-visible {
  outline: none;
  border-color: var(--vz-primary);
  box-shadow: 0 0 0 3px var(--vz-primary-tint);
}

/* A <select> needs the padding-right that its own arrow occupies. */
select.vz-control { padding-right: var(--vz-space-2); }

/* An identifier read off a screen or dictated over the phone (E3, from the auth
   layout's `.vz-input--code`). Tabular figures and tracking keep a 10-digit login id
   legible and unambiguous — a 1 and a 7, or an 8 and a B, at a counter. */
.vz-control--code {
  font-family: var(--vz-font-mono);
  font-variant-numeric: tabular-nums lining-nums;
  letter-spacing: 0.08em;
}

/* Finance's `.vz-input::placeholder` (E14). A hint inside the box is a real part of that
   product's forms — "you@company.com" under Email — and Chit's fields simply had none, so
   an empty form read as a stack of grey rectangles. `--vz-subtle` is the tertiary tone
   Finance uses, and it is the one this repo darkened in E5a so it clears 4.5 rather than
   Finance's 3.01. */
.vz-control::placeholder { color: var(--vz-subtle); opacity: 1; }

/* Finance's `.vz-input-group` + `.vz-input-addon` (E14) — the show/hide control that sits
   inside the password box. ⚠ THE PADDING ON THE INPUT IS WHAT KEEPS THE TEXT OFF THE
   BUTTON; without it a long password runs under the icon and the last characters are
   unreadable at exactly the moment somebody has pressed "show" to read them. */
.vz-control-group { position: relative; display: flex; align-items: center; }
.vz-control-group .vz-control { padding-right: 48px; }

/* ⚠ 44px, WHERE FINANCE'S ADDON IS 36. This is the one place E14 deliberately does NOT
   copy the finance portal: `--vz-tap-min` is 44 and `OPERATING_MODE §7` item 20 already
   carries two shipped controls that sit under it. Adding a third knowingly would be
   writing the next entry on that list. The input is `min-height: var(--vz-tap-min)`, so a
   44px square fills it exactly and the target is the full height of the box.

   🐛 ⚠ AND THE BUTTON ITSELF MUST NOT PAINT, WHICH IS THE BUG E14 SHIPPED. Filling a 44px
   square that sits flush at `right: 0` means the hover background covers the input's 1px
   border on three edges AND squares off its 8px corner — so hovering the reveal made the
   field stop looking like one control and start looking like a box with a grey tab stuck
   on the end. Reported by the owner; reproduced by forcing the hover state.

   The fix separates the TARGET from the PAINT. The button stays 44×44 and transparent;
   an inset pseudo-element carries the hover chip, 5px clear of the border on every side,
   with its own smaller radius. The tap target does not shrink and nothing overlaps. */
.vz-control-addon {
  position: absolute;
  right: 0;
  display: inline-flex;
  align-items: center;
  justify-content: center;
  width: var(--vz-tap-min);
  height: var(--vz-tap-min);
  padding: 0;
  border: 0;
  background: transparent;
  color: var(--vz-subtle);
  cursor: pointer;
}

/* The visible chip. Behind the icon (`z-index: -1` would escape the button's stacking
   context, so the icon is lifted instead) and inset clear of the input's border. */
.vz-control-addon::before {
  content: '';
  position: absolute;
  inset: 5px;
  border-radius: var(--vz-radius-sm);
  background: transparent;
  transition: background-color var(--vz-transition);
}

.vz-control-addon svg {
  position: relative;
  width: 20px;
  height: 20px;
  display: block;
}

.vz-control-addon:hover { color: var(--vz-ink); }
.vz-control-addon:hover::before { background: var(--vz-surface-2); }
.vz-control-addon:focus-visible { outline: 2px solid var(--vz-primary); outline-offset: -3px; }

.vz-toolbar-buttons {
  display: flex;
  align-items: center;
  gap: var(--vz-space-2);
  flex: 0 0 auto;
}

/* The money-register export bar (`§7` item 96) — a toolbar that narrows a FILE rather
   than a list, on three screens that have no list query at all.

   ⚠ THE NOTE AND THE ERROR EACH TAKE THEIR OWN LINE. `.vz-toolbar` is a wrapping flex
   row, so a `<p>` dropped into it becomes a flex item and sits BESIDE the date fields —
   at whatever width is left, which at some viewport is two words per line. These two are
   sentences, and a sentence in a flex row is `§Q-R2`'s crushed-track defect waiting to
   happen; `1 1 100%` makes each one a row of its own at every width.

   ⚠ `min-width: 0` RATHER THAN AN ENTRY IN `StylesheetStaticTest`'s ARGUED LIST, and the
   guard is what asked the question — it failed the full suite on this rule. That list is
   for growable items that must NOT shrink: a button below its label, a file input below
   "Choose file · No file chosen". **A sentence is the opposite case.** Its `min-width:
   auto` floor is its longest WORD, which is harmless in English and is exactly what an
   activated locale would break — a Tamil or German compound wider than a 320px bar would
   hold the row open with nothing able to give. Letting prose shrink and wrap is correct,
   and it keeps the argued list meaning what it says. */
.vz-toolbar--export .vz-toolbar-note,
.vz-toolbar--export .vz-field-error {
  flex: 1 1 100%;
  min-width: 0;
  margin: 0;
}

/* --- Result summary ------------------------------------------------------- */

.vz-result-bar {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  justify-content: space-between;
  gap: var(--vz-space-2);
  margin-bottom: var(--vz-space-3);
}

.vz-result-count {
  margin: 0;
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

.vz-result-count strong { color: var(--vz-ink); }

/* ⚠ ROWS PER PAGE, IN THE SPACE `.vz-result-bar` HAS BEEN HOLDING SINCE B1 (E6+ item 8).
   That bar has been `justify-content: space-between` with a single child for eleven rows;
   this is the second thing it was built for.

   ⚠ IT IS DRAWN AS THE PAGER, NOT AS A NEW COMPONENT. Same targets, same current-state
   treatment — border-width and weight, never colour alone — because it is the same idea
   one row up: "which slice of this list am I looking at". A `<select>` would have been a
   third kind of control on a screen that already has a filter bar and a pager, and it
   would have needed a form the filter bar cannot nest inside. */
.vz-per-page {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-1);
  font-size: var(--vz-fs-sm);
}

.vz-per-page-label {
  color: var(--vz-subtle);
  font-size: var(--vz-fs-xs);
  text-transform: uppercase;
  letter-spacing: 0.06em;
  margin-right: var(--vz-space-1);
}

.vz-per-page a,
.vz-per-page span.is-current {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  min-width: var(--vz-tap-min);
  min-height: var(--vz-tap-min);
  padding: 0 var(--vz-space-2);
  box-sizing: border-box;
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  color: var(--vz-ink);
  font-variant-numeric: tabular-nums lining-nums;
  text-decoration: none;
}

.vz-per-page a:hover { background: var(--vz-surface-2); }

.vz-per-page a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

.vz-per-page .is-current {
  border-color: var(--vz-primary);
  border-width: 2px;
  font-weight: var(--vz-fw-bold);
  background: var(--vz-surface-2);
}
/* --- Table ---------------------------------------------------------------- */

/* ⚠ THE SCROLL LIVES HERE, NEVER ON THE PAGE. A table wider than its column is
   the single most common cause of a horizontally scrolling page on a phone, and
   a page that scrolls sideways is unusable one-handed. Below 44rem the rows
   become cards and this never engages; between 44rem and the table's natural
   width it is the backstop. */
.vz-table-wrap {
  /* 🐛 ⚠ `position: relative` IS NOT DECORATION, IT IS THE THIRD SIGHTING OF ONE DEFECT
     (E6+ item 2). A `.vz-sr-only` span is `position: absolute`, and with no positioned
     ancestor its STATIC POSITION resolves against the initial containing block — so a
     1px hidden caption sitting in the last `<th>` of a table scrolled 696px wide
     extended the DOCUMENT to 897px inside an 805px window. The wrapper clipped the
     table perfectly and the page still scrolled sideways.

     ⚠ `.vz-step` CARRIES THE IDENTICAL COMMENT AND THE IDENTICAL FIX, written when a
     frozen step's hidden note did this inside the wizard's steps rail; `§E25` finding
     10 is the same shape a third time. Two instances were a coincidence, three are a
     rule: **anything with `overflow-x: auto` needs to be a containing block, or the
     hidden text inside it escapes to the document.**

     ⚠ FOUND ONLY BECAUSE THE DETAIL SCREENS WERE FINALLY CAPTURED. It overflowed on
     the auctions diary (770–910px, 142px) and the installment schedule (770–940px,
     174px) — inside `§E25` finding 10's band, on two screens that had never been
     measured because no fixture went deep enough to render them. */
  position: relative;
  overflow-x: auto;
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-lg);
  background: var(--vz-surface);
  /* 🐛 ⚠ AND IT OWNS THE SPACE BENEATH IT, WHICH IT NEVER DID (E6b). A table had no
     bottom margin, so anything after one butted straight against its edge —
     **measured 0px** between the reports table and the callout below it, which is
     what the owner saw. The heading pair two hundred lines up had the same defect
     and the same fix: **a component owns the room below itself**, because the thing
     that follows cannot know what it is following.

     ⚠ IT DOES NOT DOUBLE UP WITH THE PAGER. `.vz-pager`'s `margin-top` is 16px and
     adjacent margins collapse to the larger, so a paged list goes 16px → 20px rather
     than 36. Checked on all 29 captures, not assumed. */
  margin-bottom: var(--vz-space-5);
}

.vz-table {
  width: 100%;
  border-collapse: collapse;
  font-size: var(--vz-fs-sm);
}

/* ⚠ 8px VERTICAL, DOWN FROM 12 (owner, E6+ item 6) — AND THE ROW DOES NOT SHRINK BY
   TWICE THAT, BECAUSE 44px IS A FLOOR IN THE MIDDLE OF IT. `.vz-c-name a` carries
   `min-height: var(--vz-tap-min)`, which `§E25` raised deliberately after measuring what
   it cost (`§7` item 20: +8% page height on the members list, argued and accepted). So a
   members row goes 69px → 61px, not the ~52 the estimate assumed — the estimate did not
   allow for the tap target. **Shrinking that target to reach 52 would reopen a finding
   this phase closed**, so the difference is reported rather than taken out of the link. */
.vz-table th,
.vz-table td {
  padding: var(--vz-space-2) var(--vz-space-4);
  text-align: left;
  vertical-align: middle;
  border-bottom: 1px solid var(--vz-border);
}

/* 🔴 THE HEADER ROW WRAPS SINCE ITEM 63's ROW, AND THAT IS ONE RULE RATHER THAN A JUDGEMENT
   PER TABLE.
   ⏸ It carried `white-space: nowrap`, which made every column's minimum at least its own
   HEADING — so "Dividend / ticket", "Net subscription" and "What happened" each set a floor no
   content asked for. Measured at 1025 with the headers allowed to wrap and nothing else
   changed: `--statement` **17px → 0**, `--schedule` **11px → 0**, `--chits` **85 → 37**,
   `--dunning` **146 → 113**. Proved both ways — removing the override put every number back.

   ▶ **A HEADING IS A LABEL, NEVER A VALUE.** The argument for `nowrap` in this file is always
   about data somebody copies or dictates — `.vz-c-code` and `.vz-c-phone` say so in the next
   rule, and `.vz-audit-values` says *"a value split across two lines is a value somebody copies
   wrongly"*. **None of that is true of the word "Outstanding".**

   ⚠ AND IT CHANGES NOTHING BELOW 64rem, where `thead` is screen-reader-only inside the card
   layout — so this is a desktop-only relaxation of a desktop-only problem. */
.vz-table thead th {
  background: var(--vz-surface-2);
  color: var(--vz-subtle);
  font-size: var(--vz-fs-xs);
  font-weight: var(--vz-fw-bold);
  letter-spacing: 0.06em;
  text-transform: uppercase;

  /* 🔴🔴 F38: THE HEADER KEEPS ITEM 63's WRAP EVEN WHEN IT CARRIES ITS COLUMN'S CLASS.
     A figure column's header now wears the same `.vz-c-*` class its cells do, so the rule
     below aligns both — and every one of those classes also carries `white-space: nowrap`,
     which is exactly what item 63's row MEASURED and removed from this row: "Dividend /
     ticket", "Net subscription" and "What happened" each set a column floor no content asked
     for (`--statement` 17px → 0, `--schedule` 11 → 0, `--chits` 85 → 37, `--dunning` 146 →
     113). ⛔ Re-introducing it as a side effect of fixing the alignment would undo a defect
     somebody already measured twice.
     ▶ `0,1,2` beats the unprefixed `0,1,0` class rules, which is where that `nowrap` lives.
     ⚠ FOUR HEADERS STAY nowrap AND IT IS HARMLESS: the `--billing-tiers`, `--invoice-lines`
     and `--invoice-register` money rules are `0,2,0` and outrank this, and their labels are
     "Rate", "Amount", "Quantity" and "Total" — one short word each. */
  white-space: normal;
}

/* ═══ 🔴🔴 F38 — A FIGURE COLUMN ACTUALLY RIGHT-ALIGNS, WHICH IT HAS NEVER DONE ═══════════
   Raised by the owner 2026-09-24 off `/admin/billing`: a left "RATE" over a right "₹10.00".
   ⚠ MEASURING IT FOUND THE OPPOSITE DEFECT, AND A MUCH LARGER ONE.

   `.vz-table th, .vz-table td { text-align: left }` above is **`0,1,1`**, and every one of
   these classes declared its alignment unprefixed at **`0,1,0`**. ⛔ SO NONE OF IT EVER
   APPLIED. Measured 2026-09-24 against this stylesheet and against a rendered page: of the 43
   money and figure columns in the product, **4 right-aligned and 39 did not** — the four being
   the ones that happen to carry a `--modifier`-scoped rule at `0,2,0`.

   🔴 THE STYLESHEET HAS BEEN ARGUING FOR A RULE IT WAS NOT APPLYING. Two of its own comments:
   "two totals are comparable at a glance only when their last digits line up", and C3c, where
   "a minus sign wrapped away from its amount, so a negative till read POSITIVE on the next
   line". ▶ **The strongest evidence it was always meant is the card layout**: 26 rules inside
   `max-width` blocks set `text-align: left` on these very columns, which only makes sense as
   an override of a desktop right-alignment. **Both halves were written; only the half that
   needed the specificity was missing.**

   ⚠ ALIGNMENT ONLY LIVES HERE. `nowrap`, `tabular-nums`, `width` and `color` stay on each
   class's own rule at `0,1,0`, with the reasoning that belongs to them — and that is what lets
   `.vz-table thead th` above take the wrap back for headers without touching anything else.
   ⚠ AND IT IS DELIBERATELY EARLY IN THE FILE: every card override is `0,2,0` too, so it is
   SOURCE ORDER that lets them win below 64rem. Moving this block down would silently
   right-align money inside the phone cards. */
.vz-table .vz-c-money,
.vz-table .vz-c-paid,
.vz-table .vz-c-total,
.vz-table .vz-c-penalty,
.vz-table .vz-c-running,
.vz-table .vz-c-holders,
.vz-table .vz-c-hits,
.vz-table .vz-c-rate,
.vz-table .vz-c-amount,
.vz-table .vz-c-qty,
/* Pricing part 4: a payment's TDS, a second money column beside its cash (`.vz-c-paid`'s reason). */
.vz-table .vz-c-tds {
  text-align: right;
}

.vz-table tbody tr:last-child td { border-bottom: 0; }
.vz-table tbody tr:hover td { background: var(--vz-surface-2); }

/* A sortable column header. The arrow is rendered from the sort STATE, and
   `aria-sort` on the <th> carries the same fact to assistive technology — the
   direction is never signalled by the glyph alone. */
.vz-th-sort {
  display: inline-flex;
  align-items: center;
  gap: var(--vz-space-1);
  min-height: var(--vz-tap-min);
  color: var(--vz-subtle);
  text-decoration: none;
}

.vz-th-sort:hover { color: var(--vz-ink); text-decoration: underline; }

.vz-th-sort:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* 🔴 THE SORTED COLUMN WAS INVISIBLE, AND THE FIX IS `--vz-primary-strong` (owner, E6+
   item 7). `.vz-th-sort` painted `--vz-subtle` whether it was sorting or not, so the one
   column the list is ordered by looked exactly like the six that are not.

   ⚠ NEVER `--vz-primary`. The accent measures **2.59** as text and is held to a non-text
   floor on purpose; painting a word with it is `§E12`'s defect exactly, and
   `testEveryTokenAStylesheetPaintsTextWithClearsTheFloor` fails the build if it returns.

   ⚠ AND COLOUR IS NOT THE ONLY SIGNAL (WCAG 1.4.1). The `::after` arrow below says WHICH
   WAY, `aria-sort` on the `<th>` says the same thing to assistive technology, and the
   weight is `thead th`'s bold either way. The green is the third cue, not the first. */
.vz-th-sort.is-active { color: var(--vz-primary-strong); }
.vz-th-sort.is-active::after { content: '\2191'; font-size: var(--vz-fs-sm); }
.vz-th-sort.is-active.is-desc::after { content: '\2193'; }

/* How far behind a seat is — the arrears chip and its count (E6d). It wraps, unlike
   every other cell here, because the chip and the count are two elements and a narrow
   column should stack them rather than widen the table. */
.vz-c-standing { white-space: normal; }

/* The member's name — the column people scan, so it carries the weight. */
.vz-c-name { font-weight: var(--vz-fw-medium); color: var(--vz-ink); }

/* Codes and phone numbers are read digit by digit and dictated over the phone;
   tabular figures stop 1 and 7 from shifting the column. */
.vz-c-code,
.vz-c-phone {
  font-variant-numeric: tabular-nums lining-nums;
  color: var(--vz-muted);
  white-space: nowrap;
}

/* The number is a `tel:` link, because the next thing someone does after finding a
   member is call them. It therefore has to be a real target: WCAG 2.5.8 wants 24px
   minimum, and the inline-text exemption does not apply to a link sitting alone in a
   table cell. On a phone it goes to the full 44px, because that is where it is
   actually tapped. Colour is inherited so the number still reads as data; the
   underline carries the affordance. */
.vz-c-phone a {
  display: inline-flex;
  align-items: center;
  min-height: 24px;
  color: inherit;
}

.vz-c-phone a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

.vz-c-joined {
  color: var(--vz-subtle);
  white-space: nowrap;
}

/* --- Badges --------------------------------------------------------------- */

/* ⚠ THE WORD CARRIES THE MEANING; THE DOT ONLY REINFORCES IT (WCAG 1.4.1).
   Badge text is always `--vz-ink` on a faint tint, so contrast is the same
   ink-on-surface pair the theme already guarantees — a status colour is never
   asked to be readable as small text, which is a bar no green, amber or red in
   this palette clears on both themes. */
.vz-badge {
  display: inline-flex;
  align-items: center;
  gap: var(--vz-space-2);
  padding: 2px var(--vz-space-3);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface-2);
  color: var(--vz-ink);
  font-size: var(--vz-fs-xs);
  font-weight: var(--vz-fw-medium);
  /* 🔴 THE PILL MAY BREAK ACROSS TWO LINES SINCE ITEM 63's ROW, AND THE REASON IS THE SAME ONE
     `.vz-c-moved` GAVE ONE SECTION UP: an unbreakable span sets its whole COLUMN's minimum to
     its entire line. `/app/settlements` carries *"Waiting for a signature"* in one — **167px
     of pill that could not break, inside a 199px column** — and that single span was 74 of the
     96px by which the last of nineteen registers overflowed its wrap.
     ⚠ **THE DOT STAYS WITH THE FIRST LINE**, because `::before` is a flex item on a baseline
     row; what wraps is the label's own text.
     ⚠ AND THE LONG LABEL IS A COPY SMELL IN ITS OWN RIGHT — every other badge in this product
     is a word or two ("Running", "Paid", "1 missed"). Shortening it was measured too and adds
     **nothing** once the pill can wrap, so the copy is left alone rather than churned. */
  white-space: normal;
}

.vz-badge::before {
  content: '';
  width: 8px;
  height: 8px;
  border-radius: 50%;
  background: var(--vz-subtle);
  flex: 0 0 auto;
}

.vz-badge--success::before { background: var(--vz-success); }
.vz-badge--info::before    { background: var(--vz-info); }
.vz-badge--warning::before { background: var(--vz-warning); }
.vz-badge--danger::before  { background: var(--vz-danger); }

/* --- Pager ---------------------------------------------------------------- */

.vz-pager {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  justify-content: center;
  gap: var(--vz-space-1);
  margin-top: var(--vz-space-4);
}

.vz-pager a,
.vz-pager span {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  min-width: var(--vz-tap-min);
  min-height: var(--vz-tap-min);
  padding: 0 var(--vz-space-2);
  box-sizing: border-box;
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  color: var(--vz-ink);
  font-size: var(--vz-fs-sm);
  text-decoration: none;
}

.vz-pager a:hover { background: var(--vz-surface-2); }

.vz-pager a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* The current page. Weight and a solid border carry the state, not colour. */
.vz-pager .is-current {
  border-color: var(--vz-primary);
  border-width: 2px;
  font-weight: var(--vz-fw-bold);
  background: var(--vz-surface-2);
}

.vz-pager .is-disabled {
  color: var(--vz-subtle);
  background: var(--vz-surface-2);
}

/* `F93`'s reach notice, under the pager.

   ⚠ IT IS NOT A `.vz-notice`. A bordered box would read as a warning about the data, and
   nothing is wrong with the data — there is simply more of it than this control walks to.
   `.vz-form-note`'s register is the right one: subtle, extra-small, advice rather than alarm.

   ⚠ AND THE TOP MARGIN IS SMALL ON PURPOSE. `.vz-pager` already carries `margin-top: 16px`
   above itself; the sentence belongs TO the pager, so it sits close to it and takes the
   list's own bottom spacing after that. */
.vz-pager-note {
  margin: var(--vz-space-2) 0 0;
  color: var(--vz-subtle);
  font-size: var(--vz-fs-xs);
  text-align: center;
}

/* --- Empty state ---------------------------------------------------------- */

.vz-empty {
  padding: var(--vz-space-7) var(--vz-space-5);
  border: 1px dashed var(--vz-border-strong);
  border-radius: var(--vz-radius-lg);
  background: var(--vz-surface);
  text-align: center;
}

/* ⚠ `h1` BESIDE `h2` (2026-11-03): a refusal or not-found PAGE carries its title as the
   page's h1 — 25 of them had only an h2, so a screen reader's heading list began one level
   down. The same box inside a list page keeps its h2, under that page's own h1. */
.vz-empty h1,
.vz-empty h2 {
  margin: 0 0 var(--vz-space-2);
  font-size: var(--vz-fs-md);
  color: var(--vz-ink);
}

.vz-empty p {
  margin: 0 auto var(--vz-space-4);
  max-width: 32rem;
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

.vz-empty p:last-child { margin-bottom: 0; }

/* --- Notices -------------------------------------------------------------- */

.vz-notice {
  margin-bottom: var(--vz-space-4);
  padding: var(--vz-space-3) var(--vz-space-4);
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface-2);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

.vz-notice--error {
  /* Fallback FIRST: color-mix() is unsupported on Safari < 16.2, Firefox < 113
     and Chrome < 111, where the tinted declaration is discarded as invalid.
     --vz-surface-2 is defined per theme, so the degraded state stays readable
     in both rather than only in light (Phase A audit, 2026-08-24). */
  background: var(--vz-surface-2);
  background: color-mix(in srgb, var(--vz-danger) 10%, transparent);
  border-color: var(--vz-danger);
  /* Ink, not danger red: this is 13px body copy and must clear 4.5:1 on both
     themes, which the red does not. The border and tint carry the severity. */
  color: var(--vz-ink);
}

/* --- Visually hidden ------------------------------------------------------ */

.vz-sr-only {
  position: absolute;
  width: 1px;
  height: 1px;
  margin: -1px;
  padding: 0;
  overflow: hidden;
  clip: rect(0 0 0 0);
  clip-path: inset(50%);
  white-space: nowrap;
  border: 0;
}

/* --- Phones: the table becomes a stack of cards --------------------------- */

/* ⚠ NOT "hide the narrow columns". Staff on a phone are the ones standing in
   front of the member, and a column that vanishes at 375px is information they
   need and cannot get. Each row becomes a two-column card instead: nothing is
   dropped, and nothing scrolls sideways. */
@media (max-width: 64rem) {
  .vz-table-wrap {
    border: 0;
    background: none;
    overflow-x: visible;
  }

  .vz-table,
  .vz-table tbody,
  .vz-table tr,
  .vz-table td {
    display: block;
    width: auto;
  }

  /* The header row still exists for assistive technology; it simply has nothing
     to label once the grid is gone. */
  .vz-table thead {
    position: absolute;
    width: 1px;
    height: 1px;
    overflow: hidden;
    clip: rect(0 0 0 0);
    clip-path: inset(50%);
  }

  /* ⚠ THE GENERIC CARD GRID IS THE MEMBERS GRID, AND `actions` IS A NAMED AREA BECAUSE
     E6b GAVE THAT LIST A COLUMN. An unnamed cell is auto-placed in an order nobody
     chose and **still measures zero overflow, because it fits** — B5's blind spot,
     which `.vz-table--chits`, `--schedule`, `--roster` and `--worklist` each exist to
     avoid. ⚠ `/app/members` is the only table in the product rendering a bare
     `.vz-table`, so this grid serves exactly one screen; a NEW table must bring its
     own modifier rather than inherit member column names by accident. */
  .vz-table tr {
    display: grid;
    grid-template-columns: minmax(0, 1fr) auto;
    grid-template-areas:
      'name    status'
      'code    phone'
      'kyc     joined'
      'actions actions';
    gap: var(--vz-space-1) var(--vz-space-3);
    margin-bottom: var(--vz-space-3);
    padding: var(--vz-space-4);
    border: 1px solid var(--vz-border);
    border-radius: var(--vz-radius-lg);
    background: var(--vz-surface);
  }

  .vz-table tbody tr:hover td { background: none; }

  /* 🐛 🔴 `th` BESIDE `td` — `§7` ITEM 83, AND IT IS THE ONE WORD THAT ITEM NAMES.
     A `<th scope="row">` is correct markup for a row header, and `.vz-table th` gives it
     the same `border-bottom` as a data cell. Once the `tr` becomes a grid card that border
     is no longer a row separator — it is **a full-width rule drawn across the middle of the
     card**, under whichever cell happens to be the header. **Seen, not inferred**: the
     import screen drew it under every list name at 375.

     ⚠ **THE ITEM DELIBERATELY DID NOT TAKE THIS FIX**, because the cheap fix and the safe
     fix were different: one word changes **nine views**, eight of which `§Q-R5` had not
     captured, and three of those are money screens. ▶ **The difference was eight captures,
     and this row took them** — `admin/roles/_matrix`, `app/roles/_matrix`,
     `app/chits/collect`, `collect-batch`, `payout`, `app/ledger/account`,
     `ledger/transaction`, `app/members/merge-confirm` and `app/onboarding/index`, all swept
     at 320→1440 before and after. */
  /* 🐛 🔴🔴 `tbody th` — AND `§7` ITEM 83's OWN ONE-WORD FIX WOULD HAVE BEEN A REGRESSION.
     That item says the fix is *"one word in the generic block — naming `th` beside `td`"*,
     and refused to take it without the eight captures it had not made. **The captures say
     the cheap fix is wrong**, which is exactly what it suspected and could not check.

     ▶ **MEASURED AT 375 ON THE SCREENS NOBODY HAD CAPTURED:** on `app/chits/collect`,
     `app/ledger/account` and `app/ledger/transaction` the `<th scope="row">` is in
     **`<tfoot>`** — a TOTALS row — carrying `border-bottom: 1px` and `padding: 8px 16px`.
     **There it is correct**: a totals row is not a card, and its rule is the total line.
     Naming a bare `th` would have stripped the padding and the border off the totals row of
     **three money screens**.

     ⚠ **AND THE TWO ROLE MATRICES WERE NEVER AFFECTED AT ALL**:
     `.vz-table--matrix .vz-c-module` already sets `padding: 0 … ; border: 0` and outranks
     `.vz-table th`, so all 19 row headers on `/app/roles/{id}` measured 0px before this
     change. **Two of the item's nine views had their own fix already.**

     ▶ **SO THE SCOPE IS `tbody`, WHERE A ROW REALLY IS A CARD** — which is the same
     by-position reasoning `.vz-body > .vz-nav` uses one file along, and the same reason
     `.vz-table thead` is handled separately above. */
  .vz-table td,
  .vz-table tbody th {
    padding: 0;
    border-bottom: 0;
  }

  /* 🐛🔴 `tfoot` — `OPERATING_MODE §7` ITEM 93. THE LIST ABOVE NAMES `.vz-table`, `tbody`,
     `tr` and `td` AND HAS NEVER NAMED `tfoot`, so a footer inside a carded table stays
     `table-footer-group` in a `display: block` table, becomes an anonymous table box, and
     **shrink-to-fits**. The row's cells are placed correctly by whatever
     `grid-template-areas` its modifier declares — five of the six affected tables have one —
     so the placement is right and the WIDTH is wrong, which is why nobody saw it.

     ⚠ **IT MEASURES AS 0px PAGE OVERFLOW AND THE SCREENSHOT LOOKS CORRECT**, because a
     half-width card is still a card. It was found by printing the row box.

     ▶ **MEASURED ON ALL SIX AFFECTED SCREENS, both ways, at 320/375/768/1024** — the totals
     row against its own tbody row and wrap:

         --dues       328 wrap ·  245 → 328      --statement  328 wrap ·  212 → 328
         --collect    328 wrap ·  287 → 328      --legs       328 wrap ·  169 → 328
         --arrears    328 wrap ·  245 → 328      --factors    286 wrap ·   86 → 286

     ✅ **AND IT IS PROVABLY INERT WHERE IT SHOULD BE**: at 1440 every one of the six reads
     "no change" (the card block does not apply), and a control screen with no `<tfoot>` at
     all is untouched.

     🔴🔴 **THE RULE THIS ONE MUST NOT BEAT IS `.vz-table--report tfoot`**, which opts a
     statutory document back out of carding 3,300 lines below. **The two selectors have
     IDENTICAL specificity — one class, one type — so SOURCE ORDER is the only thing deciding
     it**, and that is why this rule lives here in the generic block rather than anywhere
     later in the file. ⛔ **Do not move it below line ~4340.**

     ⚠ **AND THE INJECTED SWEEP SAID THE OPPOSITE, WHICH IS THE TRAP `§Q-R5` RECORDS.** A
     `<style>` appended to `<head>` lands LAST in the cascade, so the candidate beat
     `--report` and collapsed a report's 900px totals row to 425px — a footer detached from
     the columns it totals. **Written to disk in this position it does not**, and that
     difference is the whole reason the recipe says to re-sweep from the file. */
  .vz-table tfoot { display: block; }

  .vz-c-phone a { min-height: var(--vz-tap-min); }

  .vz-c-name   { grid-area: name; font-size: var(--vz-fs-md); }
  .vz-c-status { grid-area: status; justify-self: end; }
  .vz-c-code   { grid-area: code; }
  .vz-c-phone  { grid-area: phone; justify-self: end; }
  .vz-c-kyc    { grid-area: kyc; }
  .vz-c-joined { grid-area: joined; justify-self: end; font-size: var(--vz-fs-xs); }

}

/* 🔴 SPLIT OUT OF THE BLOCK ABOVE WHEN THE CARD BREAKPOINT MOVED TO 64rem (item 63's row).
   These five are PAGE CHROME — a toolbar whose buttons go full width, a page header that
   stacks — and they are phone rules that have nothing to do with whether a table is a card.
   ⛔ They stay at 44rem deliberately: sweeping them along would make every toolbar button
   full-width on a 1000px laptop, which is not a defect anybody reported and not what the
   breakpoint move was measured for. ⚠ The ones that DID move with the cards are the ones
   coupled to them by their own comments — `.vz-action--row` (an icon-only action inside a card
   is unlabelled, and there is no hover on touch), `.vz-row-form .vz-control--inline` (a form
   in a row that has become a card) and `.vz-matrix-verb` (part of the matrix card itself). */
@media (max-width: 44rem) {
  .vz-toolbar { padding: var(--vz-space-3); }
  .vz-toolbar-buttons { flex: 1 1 100%; }
  .vz-toolbar-buttons .vz-action { flex: 1 1 auto; }
  .vz-page { align-items: stretch; }
  .vz-page-actions .vz-action { flex: 1 1 auto; }
}

/* ===========================================================================
   Forms, panels and the dialog (B2)
   ---------------------------------------------------------------------------
   The Create/Edit vocabulary. Same rules as above: no colour is introduced,
   every value is a `--vz-*` token, and nothing here needs JavaScript to be
   usable — `app.js` only opens and closes the dialog.
   =========================================================================== */

/* --- Panels: one section of a multi-section page -------------------------- */

.vz-sections { display: flex; flex-direction: column; gap: var(--vz-space-4); }

.vz-panel {
  padding: var(--vz-space-5);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-lg);
  background: var(--vz-surface);
}

.vz-panel + .vz-panel { margin-top: var(--vz-space-4); }

/* 🔴🐛 A `<form>` IS A WRAPPER AND NOT A BOX, AND THE ADJACENT-SIBLING RULE ABOVE
   CANNOT SEE THROUGH ONE (F29's drive, 2026-09-24, raised by the owner from the
   screen).

   `+` needs the two panels to be siblings. A screen that wraps its editable panel in
   a `<form>` breaks that chain TWICE: the panel INSIDE the form has no panel sibling
   before it, and the panel AFTER the form has a `<form>` before it rather than a
   panel. Both land flush against the box above — **two borders touching with no gap**,
   which reads as one thick divider with a dead strip in it.

   ⚠ MEASURED, NOT ESTIMATED: on `/admin/settings` FOUR panels sat at a 0px gap while
   the five that happen to be panel-after-panel sat at 16px, on the same screen. The
   owner found it by looking; nothing measured it, because every panel was the right
   size, in the right place, with no overflow anywhere.

   ▶ THE FORM IS MADE TRANSPARENT FOR SPACING rather than given a margin of its own —
   a margin on the `<form>` would also fire for a form that holds a single button,
   which is not a panel and does not want panel spacing.

   ⚠ AND `.vz-empty + .vz-panel` IS THE SAME OMISSION ONE ELEMENT ALONG, found by the
   same sweep on `/admin/provisioning`: an empty state is a block like any other and
   the panel after it was flush too. */
.vz-panel + form > .vz-panel,
form + .vz-panel,
form + form > .vz-panel,
.vz-empty + .vz-panel { margin-top: var(--vz-space-4); }

/* 🔴🐛 F84 — AND THE SAME OMISSION ONE ELEMENT ALONG: A HEADING AFTER A FORM.
   (2026-09-28, raised by the owner from the screen with a screenshot of
   `/app/compliance/{id}`.)

   The rule above makes a `<form>` transparent for the PANEL that follows it. A
   `.vz-panel-title` that follows one matches neither it nor `.vz-panel + .vz-panel-title`
   further down, so it lands **flush against the submit button above it** — on the filing
   screen, "Record the filing" sat directly on top of "Upload the copy" with no gap at all.

   ⚠ MEASURED, NOT ESTIMATED: **16 collisions across 9 screens**, and they are the money
   screens — `chits/payout` alone has five, `settlements/show` three, `compliance/show` two,
   plus both collect screens, both settlement wizards, the follow-up profile and the
   settlements register.

   ▶ THIS IS F29'S FINDING, ONE ELEMENT ALONG. That row wrote *"a `<form>` is a wrapper and
   not a box, and the adjacent-sibling rule cannot see through one"*, fixed it for
   `.vz-panel`, and stopped there. **The heading case was never asked about** — which is why
   it survived another nine rows and was again found by somebody looking at a screen rather
   than by anything measuring one. `PanelGapTest` now derives the population instead.

   ⚠ AND A BROWSER SWEEP OF ALL TWENTY BLOCK KINDS A VIEW PLACES AFTER A `</form>` FOUND
   THREE, not one. Measured against the same element after a `.vz-panel`, which is the shape
   that has always been right:

       .vz-empty        0px after a form · 16px after a panel   ← 23 uses, the most common
       .vz-panel-title  0px              · 16px                 ← 16 uses, the reported one
       .vz-notice       0px              · 16px
       .vz-panel-note   0px              · 16px

   ⛔ THE OTHER ELEVEN ARE 0px IN BOTH CASES and are left alone — `vz-page-actions`,
   `vz-kv`, `vz-callout` and the rest take their spacing from somewhere else, and a rule
   here would be a second opinion about them. **`.vz-empty` after a form is the one the
   owner did not report and the one that occurs most.** */
form + .vz-empty,
form + .vz-notice,
form + .vz-panel-title,

/* 🐛 F87's SECOND PAIR, AND THE ONLY OTHER THING THE 353-PAGE CRAWL FOUND UNDER 16px.
   `/app/compliance/generate` puts its count line — *"10 to create · 0 already on the
   calendar"* — after the period form, at **4px**.

   ⚠ `.vz-page-sub`'s 4px IS `--vz-space-1` AND IT IS DELIBERATE **IN ITS OWN ROLE**: the
   stylesheet records `H1 → .vz-page-sub` as *"4px on purpose and is right"*. ▶ This is the
   same class doing a DIFFERENT job — a caption for the table below it, not a subtitle for
   the heading above it — and at 4px above with `--vz-space-5` below it reads as attached to
   the form it follows rather than to the table it introduces.

   ⚠ ONE SITE IN 353 PAGES, so the blast radius is that screen; the rule cannot fire where a
   form does not directly precede, which is every page-header use of the class. */
form + .vz-page-sub,
form + .vz-panel-note { margin-top: var(--vz-space-4); }

/* 🔴🐛 F85 — AND A BARE ACTION AFTER A FIELD, WHICH IS THE SAME OMISSION INSIDE THE FORM.
   (2026-09-28, raised by the owner from the screen: *"Choose a image and save photo button
   has no space in-between"*, and the same on two compliance screens.)

   `F29` made a `<form>` transparent for the panel after it. `F84` did the heading, the empty
   state and the notice. **All three were about what follows the `</form>`** — and every case
   he reported is INSIDE one: a `<button class="vz-action">` written as a direct child rather
   than inside a `.vz-form-actions` wrapper, landing flush on the field above it.

   ⚠ MEASURED ON RENDERED MARKUP, NOT ON THE SOURCE. 117 bare actions exist across 73 views
   and **almost all of them are correct** — they sit in filter toolbars and table rows where
   no gap is wanted. Dumping all 33 parameterless screens and measuring each one in a browser
   gave 14 bare actions inside forms, of which exactly **3 are flush**, and all three share
   one predecessor: `div.vz-field`. A blanket `form > * + .vz-action` rule would have moved
   the other hundred-odd for no reason.

       compliance/generate   div.vz-field → 0px    "Show what this would create"
       compliance/new        div.vz-field → 0px    "Add the filing"
       profile               div.vz-field → 0px    "Save photo"
       …the same shape on /app/compliance/{id}, which takes a parameter and is not in the walk

   ▶ `var(--vz-space-5)` because that is what `.vz-form-actions` gives, and the point is that
   a button reads the same whether or not somebody remembered the wrapper. */
.vz-field + .vz-action { margin-top: var(--vz-space-5); }

/* 🔴🐛 AND `.vz-sections` SUPPLIES THE SPACE, SO IT TAKES THE MARGIN BACK OFF
   (owner, 2026-09-24: *"make the panel spacing one number"*).

   ⚠ THIS IS `§2b`'s SECOND HALF, WHICH THIS FILE ALREADY CITES TWELVE LINES DOWN for
   `.vz-panel > :last-child`: *"a container that supplies the space takes it back off"*.
   `.vz-sections` is a flex column with `gap: var(--vz-space-4)`, and its children are
   panels — which ALSO match the sibling rule above. **Both fire, and 16 + 16 = 32.**

   🔴 IT IS A DOUBLE-COUNT AND NOT A SECOND DESIGN. Measured across 42 screens and 68
   panels: **one** container does this, `.vz-sections`, on **30 panels across six operator
   screens** — `/app/settings`, `/app/members/{id}`, `/app/chits/{id}` and three more. Every
   other panel in the product sat at 16. ⛔ So the product had **one rhythm and one
   accident**, not two rhythms, and the accident is what the owner saw.

   ⚠ MEASURED BEFORE CHOOSING WHICH TO CANCEL: every `.vz-sections` in the product holds
   `.vz-panel` children and NOTHING else, on every screen that renders one. So taking the
   margin off inside it changes the gap between panels and can change nothing else.

   🔴 ⚠ BOTH SELECTORS TIE ON SPECIFICITY WITH THE RULES THEY OVERRIDE, SO SOURCE ORDER IS
   THE WHOLE CASCADE — `§7` item 101's lesson, which this stylesheet has now written down
   four times. `.vz-sections > .vz-panel` is (0,2,0) against `.vz-panel + .vz-panel`'s
   (0,2,0); `.vz-sections > form > .vz-panel` is (0,3,0) against `.vz-panel + form >
   .vz-panel`'s (0,3,0). ⛔ **Moving this block above either of them silently restores the
   32px**, and `StylesheetStaticTest` asserts the order for exactly that reason. */
.vz-sections > .vz-panel,
.vz-sections > form > .vz-panel { margin-top: 0; }

/* 🔴🐛 AND A PANEL'S NEIGHBOUR GETS THE GAP EVEN WHEN IT IS NOT A PANEL
   (owner, 2026-09-24, from `/admin/migrations`: *"fix the chips gap too"*).

   ⚠ **`.vz-panel` HAS NO `margin-bottom` ON PURPOSE** — panels space themselves by giving
   the NEXT panel a `margin-top`. ▶ So anything that follows a panel and is not a panel
   gets **nothing**, and lands flush against its border. Measured across 38 control-plane
   screens, five kinds of element follow a panel at the top level and **three of them sat
   at 0px**:

   | after a panel        | was | where                                             |
   |----------------------|-----|---------------------------------------------------|
   | `.vz-panel`          | 16  | 72 pairs — the rule above                          |
   | `.vz-form-actions`   | 20  | its own `margin-top`                               |
   | `.vz-notice`         | 16  | already had a rule of its own, folded in below     |
   | **`.vz-chips`**      | **0** | `/admin/migrations`                              |
   | **`.vz-panel-title`**| **0** | `/admin/users`, `/admin/audit`, `/admin/audit/sign-ins` |
   | **`.vz-panel-note`** | **0** | `/admin/organizations/new`, a connection's edit   |
   | **`.vz-empty`**      | **0** | `/app/onboarding` — found by the operator sweep   |

   ⚠ **`.vz-empty` IS THE PAIR OF `.vz-empty + .vz-panel` TWO RULES UP, THE OTHER WAY
   ROUND**, and it was found only because the sweep was run on BOTH portals: every other
   instance of this defect is on the control plane.

   ⛔ **A BLANKET `.vz-panel + *` WAS REJECTED, AND THIS FILE ALREADY ARGUES WHY** at
   `.vz-kv + .vz-panel-note`: *"a blanket sibling selector would reach the pairs that
   already have a gap and would then have two."* Measured here: it would drop
   `.vz-form-actions` from 20 to 16, and it would give a `<form>` wrapper a margin **on top
   of** the one its inner panel already gets — 32px, the double-count two rules up.

   ⚠ **SO `.vz-panel + .vz-notice` IS FOLDED IN RATHER THAN LEFT WHERE IT WAS.** It was a
   lone rule 1,100 lines away doing exactly this job for one class. One place now answers
   *"what does a panel's neighbour get"*, which is the only way the next one gets noticed. */
.vz-panel + .vz-notice,
.vz-panel + .vz-chips,
.vz-panel + .vz-empty,
.vz-panel + .vz-panel-title,
.vz-panel + .vz-panel-note { margin-top: var(--vz-space-4); }

/* 🔴🐛 AND THE PAGER HAS THE SAME OMISSION, FOUND BY DRIVING F31's NEW SCREEN.
   `.vz-pager` carries `margin-top` and no `margin-bottom` — exactly `.vz-panel`'s shape —
   so a sentence under a register lands **flush against the page buttons**.

   ⚠ MEASURED, 2026-09-24 — THREE SCREENS, AND THE THIRD IS NOT A BILLING ONE:
   | screen                      | after the pager             | was |
   |-----------------------------|-----------------------------|-----|
   | `/admin/billing/invoices`   | `.vz-panel-note`            | **0** |
   | `/admin/billing/operators`  | `.vz-panel-note` × 3        | **0** |
   | `/admin/provisioning`       | `.vz-panel.vz-panel--muted` | **0** |

   ⛔ The invoice register has shipped flush since owner's item 7 and provisioning since
   `§Q-H`; **F31 is only what made it visible**, by putting three notes where there had been
   one and their own 16px rhythm beside the missing gap.

   🔴🔴 **AND PROVISIONING NEEDED 26 ROWS TO SHOW IT.** `partials/list-pager` renders
   nothing at one page, so that screen has no pager — and therefore no defect — until the
   list pages. ⚠ **The dev database has three provisioning runs**: this was found by seeding
   thirty into the test database and looking, not by reading the view.

   ⛔ **THE SAME TWO REFUSALS AS THE RULE ABOVE.** A blanket `.vz-pager + *` would reach a
   neighbour that already has its own margin and give it two; a `margin-bottom` on
   `.vz-pager` would fire on the last element of every register, and ten of the twelve
   control-plane registers end with their pager.

   ⚠ **`.vz-panel` COVERS `--muted` AND EVERY OTHER MODIFIER**, which is why the selector is
   the base class rather than the pair that was measured.
   ⛔ **AND `<dialog class="vz-dialog">` AFTER THE PAGER ON `/app/members` IS NOT A MEMBER
   OF THIS LIST.** Measured rather than assumed: a `<dialog>` with no `open` computes
   `display: none` at height 0, so it gives no gap to miss — and an open one is a modal, out
   of flow entirely. ▶ `testAPagersNeighbourGetsTheGapToo` excludes it **by tag and says so**,
   rather than the list quietly acquiring a class that paints nothing. */
.vz-pager + .vz-panel-note,
.vz-pager + .vz-panel { margin-top: var(--vz-space-4); }

/* 🔴🐛 AND A `.vz-kv`'s NEIGHBOUR GETS ONE TOO — THE THIRD TIME THIS FAMILY HAS BITTEN, AND
   THE OWNER FOUND THIS ONE TOO (2026-09-24, from `/admin/billing`: *"'What the next invoice
   will use' has spacing issues"*, with a red box round the seam under **Days to pay**).

   `.vz-kv { margin: 0 }`, so it gives its neighbour **nothing** — `.vz-panel`'s shape and
   `.vz-pager`'s, a third time. ⚠ **AND IT HAD ALREADY BEEN PATCHED ONCE, FOR ONE CLASS,
   1,100 LINES AWAY**: `.vz-kv + .vz-panel-note` sat in the `.vz-sub-list`/`details` rule,
   which is the *exact* thing folding `.vz-panel + .vz-notice` in was supposed to stop
   happening again. ▶ **One place answers *"what does a `.vz-kv`'s neighbour get"* now.**

   ── ⚠ MEASURED BEFORE THE LIST WAS WRITTEN, AND THE LIST IS WHAT THE VIEWS ACTUALLY USE ──
   A scan over **both portals** found `.vz-kv` followed by **16 distinct classes**. Each was
   then built synthetically inside a `.vz-panel` and its computed `margin-top` read:

   | already spaced | | flush at 0px | |
   |---|---|---|---|
   | `.vz-panel-subtitle` | 20 | `.vz-kv` | **0** |
   | `.vz-form-note` | 16 | `.vz-table-wrap` (+ `--doc`) | **0** |
   | bare `<p>` | 14 | `.vz-callout` | **0** |
   | `.vz-page-actions` | 14 | `.vz-form-grid` | **0** |
   | `.vz-panel-note` | 12 (the lone patch) | `.vz-toolbar` | **0** |
   | | | `.vz-panel-empty` | **0** |
   | | | `.vz-audit-values` | **0** |
   | | | `.vz-wording` | **0** |

   🔴 **EIGHT OF SIXTEEN, AND NOT ONE OF THEM IS A PAIR WHERE 0px IS RIGHT.** The five on the
   left keep their own values and are deliberately NOT listed here — they are not flush, and
   listing them would flatten four different considered numbers to one.

   ⛔ **A BLANKET `.vz-kv + *` WAS REJECTED FOR A NEW REASON, NOT THE OLD ONE.** The rule two
   blocks up refuses a blanket because it would ADD a gap to pairs that have one; here it
   would **override** four — `.vz-panel-subtitle` 20 → 12, `.vz-form-note` 16 → 12, `<p>` and
   `.vz-page-actions` 14 → 12 — on a specificity tie broken only by source order. ▶ **Silently
   flattening a considered value is worse than doubling a gap, because nothing looks wrong.**

   ⚠ 12px (`--vz-space-3`) AND NOT 16: it is what the lone patch already gave a note, and it
   is what `.vz-result-bar` gives a `.vz-table-wrap` on every register. **The number is
   inherited from the two rules this replaces, not chosen.** */
.vz-kv + .vz-kv,
.vz-kv + .vz-table-wrap,
.vz-kv + .vz-callout,
.vz-kv + .vz-form-grid,
.vz-kv + .vz-toolbar,
.vz-kv + .vz-panel-empty,
.vz-kv + .vz-audit-values,
.vz-kv + .vz-wording,
/* Pricing part 6: the ladder's pause form follows its stage list on the billing panel; a
   `.vz-disclosure` has no margin-top of its own, so it was flush. */
.vz-kv + .vz-disclosure,
.vz-kv + .vz-panel-note { margin-top: var(--vz-space-3); }

/* 🔴🐛 F87 — AND THE THREE OTHER PREDECESSORS A `.vz-table-wrap` LANDS FLUSH AGAINST.
   (2026-09-28, the owner, with the URL: *"http://…/app/ledger/accounts/1 — check export
   this statement button and table. Issue exits in all other ledger pages"*.)

   ⚠ HE REPORTED THIS ONCE BEFORE AS "#9" AND I ANSWERED "NOT REPRODUCED". That was wrong,
   and wrong in a way worth writing down: `F85`'s sweep measures a BUTTON against its
   PREVIOUS sibling, and this defect is what FOLLOWS the `</form>`. A measurement that could
   not have seen something is not evidence of its absence.

   ⛔ AND IT IS NOT `form + .vz-table-wrap`, WHICH IS WHAT THE REPORT LOOKS LIKE. Crawling
   both portals (353 pages, links followed rather than routes listed) and measuring every
   `.vz-table-wrap` against its predecessor gives fifteen distinct pairs, of which THREE are
   flush — and only one of them involves a form:

       form                    → 0px    ledger/accounts/{id}, admin/reference-data
       .vz-toolbar-buttons     → 0px    app/audit  ("Export CSV" on the header row)
       .vz-result-count        → 0px    app/collections/followups
       ── and the eleven that are correct, which is the argument ──
       .vz-result-bar          → 12px   ×64, the registers
       .vz-panel-note          → 16px   ×39
       .vz-kv                  → 12px   the rule this joins
       .vz-report-heading      →  4px   ⛔ DELIBERATE: `margin: 0 0 var(--vz-space-1)`,
                                        a section heading tight on its own table

   ▶ **SO `F84`'s CLASSIFICATION WAS THE THING THAT WAS WRONG, NOT ITS FIX.** That row swept
   what follows a `</form>`, put eleven kinds aside as *"0px in both cases — they take their
   spacing from somewhere else"*, and `.vz-table-wrap` is **not** such an element: it is
   spaced by twelve different predecessors and flush after exactly three. It was classified
   by whether it HAS rules elsewhere rather than by what it MEASURES here.

   ⚠ NO NEW NUMBER ENTERS THE STYLESHEET. These join `.vz-kv + .vz-table-wrap`'s own
   declaration, whose comment above records `--vz-space-3` as *"what `.vz-result-bar` gives
   a `.vz-table-wrap` on every register"* — 12px, on 64 measured sites. Inherited, not
   chosen, which is this block's whole discipline.

   ⛔ AND IT IS A DECLARATION OF ITS OWN RATHER THAN THREE MORE SELECTORS ON `.vz-kv`'s.
   The first draft folded them in — same value, one rule — and turned the gate RED:
   `StylesheetStaticTest::testAKeyValueListsNeighbourGetsTheGapToo` pins that block as a
   list of `.vz-kv + …` selectors ENDING in its declaration, and a foreign selector on the
   end breaks the shape it reads. ▶ **The guard was right and the fold-in was wrong**: that
   block's comment is an argument about what `.vz-kv` does to ITS neighbours, and three
   pairs that have nothing to do with `.vz-kv` do not belong inside it. */
form + .vz-table-wrap,
.vz-toolbar-buttons + .vz-table-wrap,
/* The rate card's bands, opened from their toggle on `/app/subscription` (2026-11-15): measured
   0px under the summary in the local drive — a fourth predecessor of the same family. */
.vz-disclosure > summary + .vz-table-wrap,
.vz-result-count + .vz-table-wrap { margin-top: var(--vz-space-3); }

/* 🐛 ELEVENTH SEAM OF THE SAME FAMILY (owner, 2026-09-28, point 6: *"spacing issue in groups
   details page"*). `/app/chits/263`'s Subscribers panel reads heading → **16px** → *"Slots
   taken 20 of 20"* → **0px** → the callout explaining why nobody can be enrolled → 20px →
   the roster button. ⛔ **The explanation was welded to the figure it explains.**

   ── 🔴 THE POPULATION WAS DERIVED, NOT ASSUMED, AND THAT IS WHY IT IS ONE LINE ──
   `F87` looked like `form + .vz-table-wrap` and was three other pairs on four screens, one of
   them in the admin portal. ▶ So all **354 crawled pages** were parsed for every `.vz-callout`
   and the class of its previous ELEMENT sibling, then each distinct pair was measured in a
   browser:

       .vz-page + .vz-callout          62 sites   20px
       .vz-page-sub + .vz-callout      22         20
       .vz-callout + .vz-callout       18         16
       .vz-panel-note + .vz-callout     3         16
       .vz-panel-title + .vz-callout    2         16
       .vz-credential + .vz-callout     2         12
       .vz-table-wrap + .vz-callout     1         20
       .vz-form-grid + .vz-callout      1         20
       .vz-result-bar + .vz-callout     1         12
       .vz-facts + .vz-callout          1          0   ◀ the one he reported

   ⚠ **NINE OF TEN ALREADY HAD A GAP.** This is a single missing rule, not a family — and
   *"nine were fine"* is the half a scan that reported only the failure could not say.

   ▶ **16px IS TAKEN FROM ITS OWN NEIGHBOURS, NOT CHOSEN**: inside a panel, a callout after a
   `.vz-panel-title` and after a `.vz-panel-note` both measure 16. `.vz-facts` carries
   `margin-bottom: 0`, so the space has to come from this side. */
.vz-facts + .vz-callout { margin-top: var(--vz-space-4); }

/* 🐛 THE SAME MISSING RULE, ONE CLASS ALONG (found driving the passkeys branch, 2026-11-18).
   The operator's "Member portal" panel is a `.vz-facts` list followed by a `.vz-panel-note`,
   and the note sat flush under the last fact — "Passkeys · 1 device" touching "Issue a link
   the member can use…". It predates the branch: the list ended at "Portal password" and the
   same note followed it at 0px. 16px for `.vz-facts + .vz-callout`'s reason above. */
.vz-facts + .vz-panel-note { margin-top: var(--vz-space-4); }

/* ⚠ `§2b`'s SECOND HALF, APPLIED WHERE IT HAD NOT BEEN: *"a container that supplies the
   space takes it back off"*. Every component carries a `margin-bottom` so the thing after
   it gets its room — the rule `.vz-page .vz-page-title` already answers this way — but the
   LAST child of a panel has nothing after it, and padding stops that margin collapsing.

   Measured across 45 panels on 21 screens: a panel bottom landed on **21, 29, 37 or 41
   pixels** depending on which component happened to be last, which is not four decisions
   but one accident repeated. A panel ending in a `.vz-panel-note` sat 16px lower than the
   one above it ending in `.vz-facts`, on the same profile.

   After: **one value, 21px** (20px of padding and the border), at 1280 on every panel
   measured. ⚠ The single residual is honest and is not this rule's: at 375 the member
   profile's Seats panel reads 33, because its table is genuinely scrolling and the
   horizontal scrollbar inside `.vz-table-wrap` occupies the difference. */
.vz-panel > :last-child { margin-bottom: 0; }

.vz-panel--muted { background: var(--vz-surface-2); border-style: dashed; }

.vz-panel-title {
  margin: 0 0 var(--vz-space-4);
  font-size: var(--vz-fs-md);
  color: var(--vz-ink);
}

/* A heading for a section INSIDE a panel (E26). The settings panel now holds two ways
   to set one value — ready-made colours and your own — and they need naming without a
   second `.vz-panel-title`, which would read as a second panel. Smaller than the panel
   title, and with space above it because it opens a group rather than a card. */
.vz-panel-subtitle {
  margin: var(--vz-space-5) 0 var(--vz-space-2);
  font-size: var(--vz-fs-sm);
  font-weight: var(--vz-fw-medium);
  color: var(--vz-ink);
}

/* 🐛🔴 `:first-child`, AND IT WAS `:first-of-type` UNTIL Q4b (2026-09-10).
   The exception exists for a subtitle that IS the first thing in the panel; `:first-of-type`
   says *"the first `<h3>` among its siblings"*, and a panel's children are an `<h2>`, some
   `<p>`s, a `<ul>` and a `<table>` — so the `<h3>` is the first of ITS type almost always and
   the first CHILD almost never. **The exception therefore fired on every instance and the
   20px above a section heading was never rendered anywhere.**

   ⚠ MEASURED, NOT DERIVED: `/admin/audit`'s *"Nothing here can be changed"* reads
   `margin-top: 0px` with `matches(':first-of-type') === true` and `isFirstChild === false`.

   ⚠ **IT WAS MASKED FOR TWO ROWS BY WHATEVER PRECEDED THE HEADING.** On `/admin/audit` and
   `/admin/connections` the previous element is a `.vz-panel-note`, whose own 16px
   `margin-bottom` left a gap that looked deliberate. Q4b's run panel puts the heading after a
   `.vz-sub-list`, whose `margin-bottom` is 0 — so the gap collapsed to **exactly 0px** and the
   heading sat on the list.

   🔴 **AND `StylesheetStaticTest`'s `-of-type` GUARD EXEMPTS MARGIN RULES FOR A REASON THAT
   DOES NOT HOLD HERE**: *"`.vz-kv dt:first-of-type` means exactly what it says, because every
   child there IS a `dt`."* True there, false in a panel. The exemption is right about its
   example and wrong as a general rule. */
.vz-panel-subtitle:first-child { margin-top: 0; }

/* ⚠ RECONCILED WITH `.vz-panel-title` RATHER THAN LEFT AS A SECOND NUMBER (E6+ item 12).
   The title said 16px below itself and the note said 12px, for the same relationship —
   "a heading-ish thing, and then the content". They are both 16px now, and the pair
   `title + note` gets the SAME 4px the page header's `title + sub` gets, so one
   relationship is drawn one way in both places instead of three ways in two.

   ⚠ THE RULE BELOW THIS ONE IS A `:has()`, WHICH IS AN ENHANCEMENT. A browser without it
   renders 16px between the title and its note instead of 4 — looser, never broken —
   which is the same contract `.vz-choice:has(:checked)` holds. */
.vz-panel-note,
.vz-panel-empty {
  margin: 0 0 var(--vz-space-4);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

.vz-panel-title:has(+ .vz-panel-note) { margin-bottom: var(--vz-space-1); }

.vz-panel-empty { margin-bottom: 0; }

/* --- Facts: the read view of a profile section ---------------------------- */

/* A definition list, because that is what it is: a label and its value. Grid
   rather than the browser default indentation, so the values line up in a
   column a reader can scan straight down. */
.vz-facts {
  display: grid;
  grid-template-columns: minmax(8rem, 14rem) 1fr;
  gap: var(--vz-space-2) var(--vz-space-4);
  margin: 0;
}

.vz-facts dt { color: var(--vz-subtle); font-size: var(--vz-fs-sm); }
.vz-facts dd { margin: 0; color: var(--vz-ink); font-size: var(--vz-fs-sm); }

/* 🐛🔴 THE MEMBER PORTAL STACKS THIS GRID ON A PHONE, AND IT IS THE ONLY SURFACE THAT DOES.
   Measured on `/p/…/groups` at 320 on 2026-09-28: columns resolved to **158px + 72px**, and
   "19 of 20 seats left" wrapped into a FOUR-LINE pill.

   ⚠ **THAT IS THE DEFECT THE PORTAL HAS RECORDED TWICE AND AVOIDED TWICE.**
   `portal/home.php`: *"THIS WAS A `.vz-facts` GRID AND IT CRUSHED THE VALUE COLUMN — 58px at
   320 and 61px at 375."* `portal/profile.php`: *"`.vz-sub-list` AND NOT `.vz-facts`."*
   `portal/auctions.php` says it a third time. ⛔ **Until this rule, the open-groups page was
   the only `.vz-facts` in the whole member portal** — the one screen that walked into it.

   ▶ **STACKING IS THE FIX RATHER THAN AVOIDING THE COMPONENT**, because the reason the other
   screens avoided it was this crush, and a label above its value has neither the crush nor the
   run-on line those screens settled for. `minmax(8rem, …)` is a sound minimum on a branch
   machine and is simply more than a third of a phone.

   ⚠ **`.vz-portal`-SCOPED. THE OPERATOR APP IS UNTOUCHED** — `.vz-facts` is on dozens of its
   screens, and `F41`'s and `F82`'s band measurements were all taken against the two-column
   shape. ⚠ **30rem, NOT the 48rem the shell uses**: above 480px the value column is wide
   enough for the widest thing this page draws, and stacking a comfortable grid would waste
   the width a tablet has. */
@media (max-width: 30rem) {
  .vz-portal .vz-facts { grid-template-columns: 1fr; gap: 0 0; }
  .vz-portal .vz-facts dt { margin-top: var(--vz-space-3); }
  .vz-portal .vz-facts dt:first-of-type { margin-top: 0; }
}

/* 🐛 A FIGURE LIKE "3½ of 20" IS THREE TOKENS WHOSE MIN-CONTENT IS ONE CHARACTER, AND
   THE GRID ABOVE THEN GIVES IT ONE CHARACTER (2026-09-15).

   `1fr`'s automatic minimum is min-content, so a value that can break anywhere lets the
   LABEL track take its whole 14rem — measured at 320px: label 201px, value **14px**, and
   "0 of 20" rendered as three stacked lines. Keeping the figure on one line raises that
   floor to the figure's own width (~55px) and the label gives the space back.

   ⚠ IT IS A FIGURE, NOT A SENTENCE. `white-space: nowrap` on prose would push the whole
   line into the track's minimum — the `.vz-c-moved` trap `§8a` row 7i recorded, one
   component along — so this stays opt-in per value rather than going on `.vz-facts dd`.

   ⚠ AND IT IS NOT THE GENERAL DEFECT, WHICH IS NOT THIS ROW'S. The same grid starves
   *"The room draws for it"* to 44px on the panel below, and every `.vz-facts` in the
   product with it; that is `OPERATING_MODE §7` item 63's family and wants its own row. */
.vz-facts dd.vz-fact-figure { white-space: nowrap; }
/* A value that is a link — the tel: number — is a real target sitting alone in a
   cell, so WCAG 2.5.8s inline-text exemption does not cover it. */
.vz-facts dd a {
  display: inline-flex;
  align-items: center;
  min-height: 24px;
  color: inherit;
}

.vz-facts dd a:focus-visible { outline: 2px solid var(--vz-primary); outline-offset: 2px; }

.vz-mono { font-family: var(--vz-font-mono); letter-spacing: 0.06em; }

.vz-address {
  margin: 0;
  font-style: normal;
  color: var(--vz-ink);
  font-size: var(--vz-fs-sm);
  line-height: var(--vz-lh-base);
}

/* --- Form layout ---------------------------------------------------------- */

/* Two columns where there is room, one where there is not. `auto-fit` with a
   minimum track means the form reflows on its own rather than needing a
   breakpoint per screen size. */
/* ⚠ THE BOTTOM MARGIN IS `UI_CONVENTIONS §2b` APPLIED TO A THIRD COMPONENT (P1). That
   section — *"a component owns the room BELOW itself … the thing that follows cannot know
   what it is following, which is why the space can never be its responsibility"* — was
   written for `.vz-page` and `.vz-table-wrap`, and the grid was the next one to be caught
   by it: the staff form puts a `<fieldset>` after a field grid and measured **0px**
   between them.

   ⚠ THE BLAST RADIUS WAS MEASURED, NOT ARGUED, on the two Create forms that already ship.
   `/app/chits/new` ends with this grid and does not move. `/app/members/new` follows it
   with a `.vz-form-note` whose own `margin-top` is 16 — **adjacent margins collapse**, so
   16 against 20 resolves to 20 and that screen gains 4px. `--vz-space-5` is `§2b`s stated
   unit for exactly this relationship; a second value here is the thing that section exists
   to prevent. */

/* 🐛🔴 `min(15rem, 100%)`, NOT `15rem` — `OPERATING_MODE §7` ITEM 48, CLOSED 2026-09-15.

   `minmax(15rem, 1fr)` is a HARD floor: `auto-fit` may drop the column count to one, but
   it may not make that one column narrower than its minimum. Inside a `.vz-panel` at a
   320px viewport the available width is **231px**, so a 240px track overran by **9px** —
   and by **29px at 300**. `min(15rem, 100%)` keeps the 240px preference wherever there is
   room for it and yields to the container where there is not, which is the only part of
   the declaration that ever changes.

   ── ⚠ WHY IT WAS FILED RATHER THAN FIXED FOR NINE DAYS ────────────────────
   Nothing clips and NOTHING OVERFLOWS. Every ancestor is `overflow: visible`, the control
   still stops inside the panel's border, and the 320→1440 page sweep reads **0px at all
   1,121 widths** — the only symptom is the panel's inner padding going asymmetric. It is
   invisible to every measurement except the one that compares the INNER BOX against its
   CONTAINER, which is why P5 could only file it.

   ── ✅ THE BLAST RADIUS WAS MEASURED, NOT ARGUED (`§E6`, 42 screens, both portals) ──
   This grid draws every Create/Edit form in the product — 41 views, 69 call sites — so the
   one line owes a measurement pass, which is what made it a design row rather than a typo.
   **Before: 26 of 42 captured screens overran by exactly 9px at widths 320–328.** After:
   **0 of 42 at any width from 320 to 1440**, with every track at 1440 unchanged to the
   pixel. The change is provably inert above the point where 15rem fits, because
   `min(15rem, 100%)` IS `15rem` there — the measurement confirms the arithmetic rather
   than replacing it.

   ⚠ AND IT IS `100%`, NOT `0`. `minmax(0, 1fr)` would also stop the overrun and would
   throw away the two-column reflow this grid exists for: with no minimum, `auto-fit` packs
   as many tracks as the gap allows and a 320px form becomes two 108px columns. The floor
   has to stay a floor; it only has to stop being taller than the room. */
.vz-form-grid {
  display: grid;
  grid-template-columns: repeat(auto-fit, minmax(min(15rem, 100%), 1fr));
  gap: var(--vz-space-3) var(--vz-space-4);
  margin-bottom: var(--vz-space-5);
}

/* ⚠ THE GAP IS FINANCE'S 8px, NOT 4 (E5). The reported difference between the two
   login screens was exactly this: chit spaced label-to-input with `.vz-field-head`'s
   4px bottom margin, Finance uses an 8px flex gap on the field itself. Moving it to
   `gap` also means the hint below the control gets the same rhythm for free, which the
   margin approach could not do without a second rule. */
.vz-field { display: flex; flex-direction: column; min-width: 0; gap: var(--vz-space-2); }

.vz-field-head {
  display: flex;
  align-items: center;
  gap: var(--vz-space-1);
  /* spacing now comes from `.vz-field`'s gap (E5) */
}

.vz-label {
  font-size: var(--vz-fs-sm);
  font-weight: var(--vz-fw-medium);
  color: var(--vz-muted);
}

/* An asterisk alone is a convention, not a label. The `required` attribute and
   the error message carry the actual meaning, which is why the markup marks
   this aria-hidden — here it is purely a visual cue. */
.vz-required { color: var(--vz-danger); margin-left: 2px; }

.vz-field.is-invalid .vz-control {
  border-color: var(--vz-danger);
  border-width: 2px;
}

/* ⚠ Ink, not danger red. This is 12px text, and the red does not clear 4.5:1
   on both themes at that size. The message itself carries the meaning and the
   border only reinforces it, so a colour-blind reader loses nothing
   (WCAG 1.4.1). */
.vz-field-error {
  margin: var(--vz-space-1) 0 0;
  color: var(--vz-ink);
  font-size: var(--vz-fs-xs);
  font-weight: var(--vz-fw-medium);
}

.vz-field-error::before { content: "\26A0\FE0E"; margin-right: var(--vz-space-1); }

.vz-field-hint {
  margin: 0;
  color: var(--vz-subtle);
  font-size: var(--vz-fs-xs);
}

/* ⚠ AND IT OWNS THE ROOM BELOW ITSELF TOO (P1, `§2b`). It carried 16px above and nothing
   below, so anything placed after it sat flush against it — the staff form put a field
   grid there and measured 0px. Where a note is the last thing before `.vz-form-actions`
   the two margins collapse to 20 and nothing moves, which is why this costs the forms
   that already ship exactly nothing. */
.vz-form-note {
  margin: var(--vz-space-4) 0 var(--vz-space-5);
  color: var(--vz-subtle);
  font-size: var(--vz-fs-xs);
}

.vz-form-actions {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2);
  margin-top: var(--vz-space-5);
}

/* A long form's Save must not be a scroll away from wherever the reader is. */
.vz-form-actions--sticky {
  position: sticky;
  bottom: 0;
  padding: var(--vz-space-3) 0;
  background: var(--vz-bg);
  border-top: 1px solid var(--vz-border);
}

/* --- Checkboxes ----------------------------------------------------------- */

.vz-check {
  display: flex;
  align-items: center;
  gap: var(--vz-space-2);
  min-height: var(--vz-tap-min);
}

/* ⚠ THE OWNER REPORTED THESE AS HUGE, AND THE ANSWER IS TO SEPARATE THE INK FROM THE
   TARGET (E6+ item 10). The old rule read *"24px, not the browser default 13px and not
   20px: WCAG 2.5.8 sets 24x24 as the minimum target … a box that only MOSTLY meets the
   bar is one somebody has to re-measure later"* — and every word of that reasoning is
   kept. It just applies to the TARGET, which is now 44px, and not to the drawing, which
   is 20px. "Make the checkbox smaller" and "weaken the target" are different
   instructions and this is the first one.

   🐛 ⚠ AND `padding` WAS THE OBVIOUS WAY TO DO IT AND DOES NOT WORK. A native checkbox is
   a replaced element: browsers ignore padding on it and scale the drawn tick to fill
   whatever `width`/`height` say, which is exactly why 24px "reads as huge" and why 44px
   would read as enormous. The only way to hold a 44px target around a 20px box is to
   stop the browser drawing it — `appearance: none` — and draw both halves here.

   ⚠ THE NEGATIVE MARGINS ARE ARITHMETIC, NOT A NUDGE. The target is 12px wider than the
   ink on each side; `-12px` on both sides puts the ink's left edge exactly where the old
   24px box's was, and leaves the label exactly one `gap` from the ink. So nothing moves
   on any screen — the drawing shrinks 24 → 20 and the target grows 24 → 44 underneath
   it. What the target overlaps on the right is `.vz-check-label`, which activates the
   SAME control, so it steals no clicks from anything; on the left it reaches into the
   form's own gutter, where there is nothing to steal from. That is `§E25`'s argument for
   `.vz-c-name a`, applied to the other end of the same rule.

   ⚠ ONE RULE DRAWS BOTH A CHECKBOX AND A RADIO, because one class already names both
   (`branch.php`, `nominees.php`, the collections tick column). The shape is the only
   difference: a square with a tick, or a circle with a dot.

   ⚠ AND `accent-color` IS GONE BECAUSE IT NO LONGER APPLIES — it colours the control the
   browser draws, and the browser has stopped drawing one. `--vz-primary-button` replaces
   it as the fill, which is the token that is MEASURED against its own label
   (`--vz-primary-contrast`, 4.96) and re-derived per organization by `BrandPalette`. */
.vz-check-input {
  appearance: none;
  -webkit-appearance: none;
  box-sizing: border-box;
  display: grid;
  place-items: center;
  flex: 0 0 auto;
  width: var(--vz-tap-min);
  height: var(--vz-tap-min);
  margin: 0 calc(-1 * var(--vz-space-3));
  padding: 0;
  border: 0;
  background: none;
  cursor: pointer;
}

/* THE INK — a 20px box centred inside the 44px target. Both pseudo-elements sit in the
   same grid cell so the tick can be drawn over the fill without a second element. */
.vz-check-input::before,
.vz-check-input::after { content: ''; grid-area: 1 / 1; }

.vz-check-input::before {
  box-sizing: border-box;
  width: 20px;
  height: 20px;
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-sm);
  background: var(--vz-surface);
  transition: background-color var(--vz-transition), border-color var(--vz-transition);
}

.vz-check-input[type='radio']::before { border-radius: 50%; }

.vz-check-input:hover::before { border-color: var(--vz-primary); }

.vz-check-input:checked::before {
  border-color: var(--vz-primary-button);
  background: var(--vz-primary-button);
}

/* ⚠ THE TICK IS A MASK, NOT AN IMAGE, FOR `--vz-primary-contrast`'s SAKE. A colour
   cannot be interpolated into a data URI, and the label colour on the fill is a token an
   organization's palette can move. The path is `partials/icon.php`'s `check`, copied at
   the moment it moved here — the same argument E23's callout glyphs record. */
.vz-check-input::after {
  width: 14px;
  height: 14px;
  background: var(--vz-primary-contrast);
  -webkit-mask-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='%23000' stroke-width='3' stroke-linecap='round' stroke-linejoin='round'%3E%3Cpath d='m5 12 5 5L20 7'/%3E%3C/svg%3E");
          mask-image: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='%23000' stroke-width='3' stroke-linecap='round' stroke-linejoin='round'%3E%3Cpath d='m5 12 5 5L20 7'/%3E%3C/svg%3E");
  -webkit-mask-repeat: no-repeat;
          mask-repeat: no-repeat;
  -webkit-mask-size: contain;
          mask-size: contain;
  opacity: 0;
}

/* A radio has no tick: the same node becomes the dot. */
.vz-check-input[type='radio']::after {
  width: 8px;
  height: 8px;
  border-radius: 50%;
  -webkit-mask-image: none;
          mask-image: none;
}

.vz-check-input:checked::after { opacity: 1; }

/* ⚠ WRITTEN BECAUSE A CUSTOM CONTROL LOSES THE PLATFORM'S OWN DISABLED RENDERING.
   Nothing in the product renders a disabled tick today; with `appearance: none` one
   would be indistinguishable from a live control the moment something does, and that is
   a different thing from the dead CSS item 3 of this row is about — this rule fires
   whenever the state it names exists. */
.vz-check-input:disabled { cursor: not-allowed; }
.vz-check-input:disabled::before { opacity: 0.55; }

/* The label is as much the tap target as the box is — a 20px checkbox on its
   own is well under the minimum. */
.vz-check-label {
  display: inline-flex;
  align-items: center;
  min-height: var(--vz-tap-min);
  color: var(--vz-ink);
  font-size: var(--vz-fs-sm);
  cursor: pointer;
}

/* ⚠ THE OFFSET IS NEGATIVE SO THE RING SITS ON THE INK, NOT ON THE 44px TARGET — a 48px
   ring around a 20px box reads as belonging to something else. The COLOUR and the WIDTH
   are the two things `testEveryFocusVisibleRuleDrawsTheSameRing` holds identical across
   all 29 rings; the offset is per-component geometry and that test deliberately does not
   assert it. -10px puts it 2px clear of the box. */
.vz-check-input:focus-visible { outline: 2px solid var(--vz-primary); outline-offset: -10px; }

/* ===========================================================================
   The searchable select (E18) — `vazhai-finance`'s `.vz-combo`, ported
   ---------------------------------------------------------------------------
   ⚠ EVERY RULE HERE IS DEAD UNLESS `app.js` RAN. Nothing in the product renders
   these class names server-side: the script builds them around a real `<select>`
   that stays the form value. Scripts off and this whole block matches nothing,
   which is the same contract the reveal button and the dialogs hold.

   The class names are Chit's (`vz-combo-*`) rather than Finance's BEM-ish
   `vz-combo__*`, because 2,900 lines of this file use single hyphens and one
   component spelling itself differently is the inconsistency this phase exists
   to remove.
   =========================================================================== */
.vz-combo { position: relative; }

/* The trigger wears `.vz-control`, so it inherits the input's height, border,
   radius, focus ring and invalid state — one control, not a lookalike. */
.vz-combo-btn {
  display: flex;
  align-items: center;
  justify-content: space-between;
  gap: var(--vz-space-2);
  width: 100%;
  text-align: left;
  cursor: pointer;
}

.vz-combo-label { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }

/* An unchosen select reads as a placeholder, exactly as an empty input does. */
.vz-combo-btn.is-placeholder .vz-combo-label { color: var(--vz-subtle); }

/* ⚠ NO `transition` ON THE ROTATION. E13 fixed the motion layer as COLOUR ONLY;
   animating a transform would extend that decision rather than apply it. */
.vz-combo-caret { flex: none; color: var(--vz-subtle); }
.vz-combo.is-open .vz-combo-caret { transform: rotate(180deg); }

/* ⚠ `fixed`, NOT `absolute`, AND THE SCRIPT POSITIONS IT. A dropdown opened
   inside `.vz-table-wrap` — which sets `overflow-x: auto` — would otherwise be
   clipped by its own scroll container, and the panel would be unreachable on the
   money screens where the long lists actually live. */
.vz-combo-panel {
  position: fixed;
  z-index: 1100;
  display: flex;
  flex-direction: column;
  /* 320 rather than Finance's 280: a 44px row is taller than its 36px one, and this
     keeps SEVEN options visible instead of four. */
  max-height: 320px;
  max-width: min(22rem, calc(100vw - 16px));
  overflow: hidden;
  background: var(--vz-surface);
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  box-shadow: var(--vz-shadow-md);
}

.vz-combo-panel[hidden] { display: none; }

.vz-combo-search-wrap {
  padding: var(--vz-space-2);
  border-bottom: 1px solid var(--vz-border);
}

.vz-combo-search-wrap[hidden] { display: none; }

/* 🐛 ⚠ THIS RULE EXISTS BECAUSE A TEST DEMANDED IT, AND THE TEST WAS RIGHT. The search
   box wears `.vz-control`, so it rendered correctly with no rule of its own — which is
   precisely the shape of every unstyled-class defect this repo has shipped: it looks
   fine, measures fine, and the hook is a lie. `testEveryClassAppJsBuildsHasACssRule`
   failed on it the first time it ran.

   What it actually needs is a smaller type size than a form field: this is a filter
   inside a 22rem popover, not an input somebody fills in. */
.vz-combo-search {
  width: 100%;
  font-size: var(--vz-fs-sm);
}

.vz-combo-list {
  list-style: none;
  margin: 0;
  padding: var(--vz-space-1);
  overflow-y: auto;
}

/* ⚠ 44px, WHERE FINANCE'S OPTION IS ABOUT 36. An option in a dropdown is a tap
   target like any other, and `OPERATING_MODE §7` item 20 already carries two
   shipped controls under `--vz-tap-min`. This is E14's call about the password
   addon, applied to the control that has the most of them on screen at once. */
.vz-combo-opt {
  display: flex;
  align-items: center;
  gap: var(--vz-space-2);
  min-height: var(--vz-tap-min);
  /* ⚠ NO VERTICAL PADDING — `min-height` AND padding COMPOUND. With `space-2` top and
     bottom the row measured 60px, so a 280px panel showed four options and the list
     read as loose rather than as a list. `align-items: center` already centres the
     label, so the floor is the height. */
  padding: 0 var(--vz-space-3);
  border-radius: var(--vz-radius-sm);
  cursor: pointer;
  font-size: var(--vz-fs-sm);
  color: var(--vz-ink);
}

.vz-combo-opt[hidden] { display: none; }

/* `is-active` is the KEYBOARD highlight and `:hover` is the pointer one; they
   look the same on purpose, because they mean the same thing. */
.vz-combo-opt:hover,
.vz-combo-opt.is-active { background: var(--vz-surface-2); }

/* 🐛 ⚠ FINANCE PAINTS THIS `--vz-primary-dark`, WHICH MEASURES 4.34 ON THE PANEL
   AND 3.83 ON A HOVERED ROW — under AA, both. That is exactly E12's defect
   (`--vz-primary` as a green WORD) living in the source this row is porting from.
   `--vz-primary-strong` is the token E12 established for a green word: 5.20 and
   4.58. ⚠ `testEveryTokenAStylesheetPaintsTextWithClearsTheFloor` is what would
   have caught it had this been copied across unread. */
.vz-combo-opt[aria-selected='true'] {
  color: var(--vz-primary-strong);
  font-weight: var(--vz-fw-medium);
}

.vz-combo-opt-text { flex: 1 1 auto; min-width: 0; overflow: hidden; text-overflow: ellipsis; }

/* The tick reserves its space so option text stays aligned whether or not the
   row is chosen — an icon that appears and shifts the label is a list that
   jitters as you pick down it.

   🐛 ⚠ AND IT IS `--vz-primary-strong`, NOT FINANCE'S `--vz-primary`. E12's general
   assertion caught this the first time it ran over the new rules: the accent measures
   **2.39** on the panel, which is under the 4.5 text floor AND under the 3.0 non-text
   floor for a meaningful graphic (WCAG 1.4.11). On a multi-select the tick is the
   primary signal that a row is chosen, so a 2.39 tick is the state being nearly
   invisible. `--vz-primary-strong` measures 5.20 and matches the label beside it. */
.vz-combo-check { flex: none; color: var(--vz-primary-strong); opacity: 0; }
.vz-combo-opt.is-checked .vz-combo-check { opacity: 1; }

.vz-combo-empty {
  padding: var(--vz-space-3);
  color: var(--vz-subtle);
  font-size: var(--vz-fs-sm);
  text-align: center;
}

.vz-combo-empty[hidden] { display: none; }

/* --- Search, add, remove: the audience picker ------------------------------ */

.vz-picker {
  min-width: 0;
  margin: var(--vz-space-4) 0 0;
  padding: 0;
  border: 0;
}

.vz-picker[hidden] { display: none; }

.vz-picker-box { position: relative; }

.vz-picker-box .vz-control { width: 100%; }

.vz-picker-panel {
  position: absolute;
  z-index: 20;
  top: 100%;
  left: 0;
  right: 0;
  max-height: 320px;
  margin-top: var(--vz-space-1);
  overflow-y: auto;
  background: var(--vz-surface);
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  box-shadow: var(--vz-shadow-md);
}

.vz-picker-panel[hidden] { display: none; }

.vz-picker-chosen {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2);
  margin: var(--vz-space-3) 0 0;
  padding: 0;
  list-style: none;
}

.vz-picker-chosen[hidden] { display: none; }

/* ⚠ A TAG MAY WRAP: `.vz-chip` is `nowrap`, and an operator's name is as long as they chose. TWO
   CLASSES, because `.vz-chip` is declared later in this file and one would lose to it: a long name
   pushed the page to 532px on a phone. */
.vz-chip.vz-picker-chip {
  max-width: 100%;
  white-space: normal;
  color: var(--vz-ink);
}

.vz-chip.vz-picker-chip::before { content: none; }

/* The same long name, in the results. */
.vz-picker-panel .vz-combo-opt { flex-wrap: wrap; padding-block: var(--vz-space-1); overflow-wrap: anywhere; }

.vz-picker-chip-text { min-width: 0; overflow-wrap: anywhere; }

.vz-picker-remove {
  flex: none;
  min-width: var(--vz-tap-min);
  min-height: var(--vz-tap-min);
  margin: calc(var(--vz-space-2) * -1) calc(var(--vz-space-2) * -1) calc(var(--vz-space-2) * -1) 0;
  padding: 0;
  border: 0;
  border-radius: var(--vz-radius-pill);
  background: transparent;
  color: var(--vz-muted);
  font-size: var(--vz-fs-lg);
  line-height: 1;
  cursor: pointer;
}

.vz-picker-remove:hover { color: var(--vz-ink); }

.vz-picker-foot {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-3);
  margin: var(--vz-space-2) 0 0;
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

/* --- Date fields: native, restyled as far as CSS allows (E24) ------------- */

/* ⚠ THE OWNER REVERSED THEIR OWN REQUEST HERE, ON THE EVIDENCE. They asked for a
   "modern, polished calendar/date-picker UI"; shown that `vazhai-finance` uses native
   `<input type="date">` in ALL FOURTEEN of its date fields — so a custom calendar would
   be Chit diverging from Finance two rows after asking the two products to look the
   same — and that a calendar is the hardest widget in the set to get right for keyboard
   and screen readers, they chose native, restyled.

   ⚠ AND MOST OF THAT WAS ALREADY TRUE. Measured before anything was written: a date
   input and a text input already agree on font, size, colour, height, border, radius and
   padding, because both wear `.vz-control`. The field, its border and its focus ring
   take the theme today. **What did not was the calendar indicator** — the browser's own
   dark glyph, with `cursor: default` on a control that is unmistakably clickable.

   ⚠ WHAT THIS HONESTLY BUYS AND DOES NOT BUY. The field and the indicator are ours; the
   DROPDOWN is the operating system's and cannot be styled, so a Chrome user still gets
   Chrome's calendar. That limitation is the reason a custom picker keeps being proposed,
   and it is accepted here rather than hidden. */

:root { --vz-calendar-glyph: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='%23000' stroke-width='2' stroke-linecap='round' stroke-linejoin='round'%3E%3Crect x='3' y='5' width='18' height='16' rx='2'/%3E%3Cpath d='M8 3v4'/%3E%3Cpath d='M16 3v4'/%3E%3Cpath d='M3 11h18'/%3E%3C/svg%3E"); }

/* The glyph, drawn here for the same reason the four alert glyphs are: nothing renders
   it as markup, so a copy in `partials/icon.php` would be a second definition of one
   drawing — and `testEveryGlyphIsRenderedBySomething` would call it an orphan. */
.vz-control[type='date']::-webkit-calendar-picker-indicator,
.vz-control[type='datetime-local']::-webkit-calendar-picker-indicator,
.vz-control[type='month']::-webkit-calendar-picker-indicator {
  /* A real target rather than the ~12px default: this is a control somebody taps on a
     phone with one hand. It cannot reach `--vz-tap-min` without pushing the date text
     off-centre, so it takes WCAG 2.5.8's 24px floor, which is the same compromise
     `.vz-check-input` records. */
  width: 24px;
  height: 24px;
  padding: 0;
  margin-left: var(--vz-space-2);
  cursor: pointer;

  /* The default is an opaque dark glyph that ignores the palette. Masking `currentColor`
     is what makes it follow the field's own ink, exactly as the callout icons do. */
  background: var(--vz-subtle);
  -webkit-mask-image: var(--vz-calendar-glyph);
          mask-image: var(--vz-calendar-glyph);
  -webkit-mask-repeat: no-repeat;
          mask-repeat: no-repeat;
  -webkit-mask-position: center;
          mask-position: center;
  -webkit-mask-size: 20px 20px;
          mask-size: 20px 20px;
}

.vz-control[type='date']:hover::-webkit-calendar-picker-indicator,
.vz-control[type='date']:focus::-webkit-calendar-picker-indicator { background: var(--vz-ink); }

/* An invalid field's indicator follows its border, so the state reads on the whole
   control rather than on three of its four edges. */
.vz-control[aria-invalid='true']::-webkit-calendar-picker-indicator { background: var(--vz-danger-text); }

/* --- A boxed choice list (E16) -------------------------------------------- */

/* ⚠ THIS IS THE BRANCH PICKER'S SIX INLINE `style` ATTRIBUTES, GIVEN A NAME. One of
   them sat INSIDE the `foreach`, so an operator with ten branches shipped ten copies of
   the same nine declarations — the cost of an inline style is per ELEMENT, not per
   template, which is the part a source-file count of 18 does not show.

   ⚠ AND IT IS BUILT ON `.vz-check-input`, NOT BESIDE IT. The product already draws
   radios one way (`members/nominees.php`, the collections tick column): 24px, WCAG
   2.5.8's floor, `accent-color` from the palette. A second radio treatment on the one
   screen a user meets before anything else is how two of them end up in the product. */
.vz-choices {
  /* A `<fieldset>` arrives with a border, padding and margins from the user agent, and
     what this element is FOR is grouping the radios for a screen reader. The legend
     carries the question; the rows carry the frame. */
  border: 0;
  padding: 0;
  margin: 0 0 var(--vz-space-5);
}

.vz-choices > legend { padding: 0; }

/* The whole row is the `<label>`, so the target is the box rather than the dot in it. */
.vz-choice {
  display: flex;
  align-items: center;
  gap: var(--vz-space-3);
  min-height: var(--vz-tap-min);
  padding: var(--vz-space-2) var(--vz-space-3);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  margin-bottom: var(--vz-space-2);
  background: var(--vz-surface);
  cursor: pointer;
  transition: background-color var(--vz-transition), border-color var(--vz-transition);
}

/* ⚠ HERE THE ROW IS THE TARGET, SO THE INPUT KEEPS ONLY ITS INK (E6+ item 10). The rule
   above already says it — *"the whole row is the `<label>`, so the target is the box
   rather than the dot in it"* — and a 44px control inside a row that pads 8px would make
   the row 60px tall to hold a target the row already provides. The 20px box is the
   base rule's `::before` either way, so this is the only screen in the product where the
   two halves of that rule are drawn by different elements. */
.vz-choice .vz-check-input {
  width: 20px;
  height: 20px;
  margin: 0;
}

.vz-choice:last-child { margin-bottom: 0; }
.vz-choice:hover { background: var(--vz-surface-2); }

/* ⚠ THE CHECKED ROW'S BORDER IS AN ENHANCEMENT, NOT THE STATE ITSELF. `:has()` is
   widely supported and unsupported browsers lose nothing that matters: the radio's own
   dot is still what says which row is chosen, and colour is never the only signal
   (WCAG 1.4.1). `--vz-primary` is the accent here rather than text, which is the role
   `theme.css` holds it to. */
.vz-choice:has(:checked) { border-color: var(--vz-primary); }

.vz-choice-label { color: var(--vz-ink); font-size: var(--vz-fs-sm); }

/* --- Field help (UI_CONVENTIONS §1) --------------------------------------- */

/* `<details>` is the whole mechanism: keyboard-focusable, toggles on Enter and
   Space, and opens on TAP rather than hover — which is §1's "mobile = tap to
   open" requirement met by the platform instead of by a script. */
.vz-help { position: relative; display: inline-flex; }

/* ⚠ THE OWNER REPORTED THESE AS TOO BIG, AND WHAT SHRINKS IS THE INK (E6+ item 13).
   The box is 24px + a 1px border = 26px, and `§E25` argued that number in BOTH
   directions: *"a 44px hit area would overlap the label by 5px and the input by 1px —
   that steals clicks from two primary controls to enlarge a secondary one"*. It cannot go
   under 24 either; that is WCAG 2.5.8's minimum target, the same floor `.vz-check-input`
   cites two hundred lines up.

   ✅ WHAT MADE IT READ AS HEAVY WAS NEVER ITS SIZE. A filled `--vz-surface-2` disc with a
   `--vz-border-strong` ring and a BOLD mono glyph, sitting beside a regular-weight label:
   three pieces of emphasis on the least important thing in the row. The fill and the ring
   go, the glyph drops to normal weight, and **both come back on hover and focus** — where
   they are an affordance rather than noise.

   ⚠ THE BORDER STAYS AT 1px AND TURNS TRANSPARENT rather than being removed, so the box
   is 26px in both states and nothing on the row moves when you point at it.

   ⚠ THIRD INSTANCE OF ONE PRINCIPLE IN THIS ROW — the checkbox (item 10) and `§E25`'s row
   link are the other two: **when a control looks too big, shrink the ink, never the
   target.** */
/* 🔄 ⚠ E6+ TOOK THE DISC AWAY AND THE OWNER PUT IT BACK, AND BOTH ARE RIGHT (E6b).
   The report was *"the help icons are too big"*; E6+ answered it by removing the fill and
   the ring, which left a **bare `i` beside a label** — *"it looks a bit odd"*, and it is:
   a lone letter in a form reads as a typo rather than as a control.

   ✅ SO THE DISC COMES BACK AND THE **DRAWING** SHRINKS INSTEAD — 26px → 18px, a third
   smaller — while the TARGET stays 24px, which is WCAG 2.5.8's floor and the number
   `§E25` argued in both directions. **This is the same move as the checkbox and the row
   link**: when a control looks too big, shrink the ink, never the target. The row that
   introduced that rule had two instances of it; this is the third, applied to the control
   that prompted the rule in the first place.

   ⚠ THE ELEMENT ITSELF NOW HAS NO BORDER, SO THE BOX IS 24px RATHER THAN 26. The ring is
   `::before`, inset 3px — which is what lets the ink and the target differ at all. */
.vz-help-toggle {
  position: relative;
  display: inline-flex;
  align-items: center;
  justify-content: center;
  width: 24px;
  height: 24px;
  border: 0;
  background: none;
  color: var(--vz-muted);
  font-family: var(--vz-font-mono);
  font-size: var(--vz-fs-xs);
  font-weight: var(--vz-fw-regular);
  cursor: help;
  list-style: none;
  user-select: none;
}

/* The disc: 18px, centred in the 24px target. */
.vz-help-toggle::before {
  content: '';
  position: absolute;
  inset: 3px;
  border-radius: 50%;
  border: 1px solid var(--vz-border-strong);
  background: var(--vz-surface-2);
  transition: background-color var(--vz-transition), border-color var(--vz-transition);
}

/* ⚠ LIFTED ABOVE THE DISC, WHICH IS WHY `FieldHelp` WRAPS THE LETTER AT ALL. A
   positioned pseudo-element paints over its parent's text, and a bare text node has no
   selector to lift it back. `.vz-control-addon svg` records the same trick. */
.vz-help-glyph { position: relative; line-height: 1; }

/* The default disclosure triangle would sit beside the glyph. */
.vz-help-toggle::-webkit-details-marker { display: none; }
.vz-help-toggle::marker { content: ""; }

/* The affordance. `[open]` is in the list because a popover that is showing should look
   like the control that opened it.

   ⚠ `:focus`, NOT `:focus-visible`, AND THE GUARD IS WHY. This is an emphasis treatment;
   the focus INDICATOR is the rule below it, and `testEveryFocusVisibleRuleDrawsTheSameRing`
   holds every `:focus-visible` selector in the product to one outline. Grouping the two
   made this selector look like a focus indicator that had lost its colour — a WCAG 2.4.7
   failure by the letter of the test — when it is a hover treatment that happens to include
   the keyboard. */
.vz-help-toggle:hover,
.vz-help-toggle:focus,
.vz-help[open] > .vz-help-toggle { color: var(--vz-ink); }

.vz-help-toggle:hover::before,
.vz-help-toggle:focus::before,
.vz-help[open] > .vz-help-toggle::before {
  border-color: var(--vz-muted);
  background: var(--vz-surface);
}

.vz-help-toggle:focus-visible { outline: 2px solid var(--vz-primary); outline-offset: 2px; }

/* 🐛 ⚠ `OPERATING_MODE §7` ITEM 19, CLOSED — AND THE FIX IS THE CONTAINING BLOCK, NOT A
   WIDTH (E6+ item 1). The popover pushed the page sideways between roughly 704 and 830px,
   and `§E25` finding 12 had already recorded what the cause is NOT: `box-sizing` shaves
   24px off a 352px panel and leaves the defect standing.

   The cause is that this box resolved against the ⓘ. `.vz-help` was `position: relative`,
   so a 22rem panel hung off a 24px icon and had nothing to be bounded BY — `max-width`
   could only ever be a guess about how much room happened to be to the right of it, and
   `70vw` is a measurement of the WINDOW, which knows nothing about the column the field
   is in. Above 704px the guess was wrong; below it, the old media query switched the
   containing block to the field and the defect vanished. **The phone rule was the fix all
   along, applied to one third of the width range.**

   ✅ SO IT IS THE FIELD AT EVERY WIDTH. `.vz-field` is a real layout box that already
   fits the page — that is what `min-width: 0` on `.vz-main` guarantees since `§E25`
   finding 10 — so `max-width: 100%` of it CANNOT extend past the document, at any
   viewport, by construction rather than by measurement. Swept 320→1440 with every
   popover forced open; the old rule overflowed on the member form from 704 to 830 and
   this one overflows nowhere.

   ⚠ AND `box-sizing` IS LOAD-BEARING NOW, WHERE IT WAS IRRELEVANT BEFORE. `max-width:
   100%` plus 12px of padding on each side is 100% + 24px without it — the popover would
   overhang its own field by exactly the padding. It was on this rule's exception list in
   `testEveryRuleThatSizesAndPadsSaysWhichBoxItMeans` with the reason "box-sizing is not
   the cause"; that reason was true of the old rule and false of this one, so the entry
   is deleted rather than left to describe a stylesheet that has changed. */
.vz-help-body {
  position: absolute;
  z-index: 20;
  top: calc(100% + var(--vz-space-1));
  left: 0;
  width: max-content;
  max-width: 100%;
  box-sizing: border-box;
  padding: var(--vz-space-3);
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  box-shadow: var(--vz-shadow-md);
  color: var(--vz-ink);
  font-size: var(--vz-fs-xs);
  font-weight: var(--vz-fw-regular);
  line-height: var(--vz-lh-base);
  text-align: left;
  white-space: normal;
}

/* ⚠ THE DECLARATIONS THAT MAKE A REAL LAYOUT BOX THE CONTAINING BLOCK, AND THEY ARE NO
   LONGER INSIDE A MEDIA QUERY (E6+ item 1). They used to be the phone's special case; the
   comment above says why they are now the rule.

   🐛 ⚠ AND E6+ ANCHORED THEM TO THE WRONG BOX (E6b). `.vz-field` is the whole field —
   label row, control, hint — so `top: 100%` put the popover **below the input**, which
   the owner reported as *"displayed way below, not near the icon"*, and made it look
   like a different component in a dialog where the field is wider. The ROW the ⓘ
   actually sits in is `.vz-field-head` (or `.vz-check` for a checkbox), which is the
   same width as the field and one line tall — so the popover lands **directly under the
   icon** and the bound that stops it leaving the page is unchanged. Proximity was the
   defect; the guarantee was never in question.

   ⚠ `.vz-field` KEEPS ITS `position` AS THE BACKSTOP. A future surface that renders a ⓘ
   straight into a field with no head row still gets a containing block that fits the
   page, rather than falling back to the initial one and overflowing the document. */
.vz-field { position: relative; }
.vz-field-head,
.vz-check { position: relative; }
.vz-help { position: static; }

/* --- Notices -------------------------------------------------------------- */

.vz-notice--warn {
  /* Fallback first, as everywhere else: color-mix() is unsupported below
     Safari 16.2 / Firefox 113 / Chrome 111 and the declaration is discarded. */
  background: var(--vz-surface-2);
  background: color-mix(in srgb, var(--vz-warning) 12%, transparent);
  border-color: var(--vz-warning);
  /* Ink, not amber: body copy at 13px, and the warning colour does not clear
     4.5:1 on both themes. The border and tint carry the severity. */
  color: var(--vz-ink);
}

.vz-notice-title { margin: 0 0 var(--vz-space-2); font-weight: var(--vz-fw-bold); }
.vz-notice p:last-child { margin-bottom: 0; }
.vz-notice-list { margin: 0 0 var(--vz-space-2); padding-left: var(--vz-space-5); }

/* ⚠ A LINK IN A NOTICE IS STILL A TAP TARGET (WCAG 2.5.8). The duplicate-member warning
   puts one on the member profile, and at a phone width it renders 17px high — the
   inline-text exemption does not cover a link that IS the whole list item. Found by
   MEASURING the profile with duplicates actually present, which is a state B2's own pass
   never produced. */
.vz-notice-list a {
  display: inline-flex;
  align-items: center;
  min-height: 24px;
  color: inherit;
}

/* --- Dialog (PHASE_B §B-JS) ----------------------------------------------- */

/* 🐛 ⚠ `box-sizing` WAS MISSING AND IT PUT EVERY MODAL 40px OFF A PHONE SCREEN
   (`§E25` finding 12, found while building E27). This stylesheet declares the property
   per component rather than in a universal reset — `.vz-action` does, `.vz-control`
   does — and this rule never did. With `content-box`, the mobile sheet's `width: 100vw`
   became 100vw PLUS `var(--vz-space-5)` of padding on each side: **415px inside a 375px
   viewport**, so the close ✕ that `.vz-dialog-head` pushes to the right edge sat
   **20px off-screen**, on every Create/Edit modal in the product since B2.

   ⚠ AND NO OVERFLOW SWEEP COULD SEE IT. A modal `<dialog>` renders in the TOP LAYER, so
   it never pushes the document: `scrollWidth - clientWidth` reads 0 while the box hangs
   off the side. `§E25` finding 10's band sweep measured these pages and was right about
   all of them. It is the same root cause as `.vz-menu-item`'s, one row earlier, and
   `StylesheetStaticTest::testEveryRuleThatSizesAndPadsSaysWhichBoxItMeans` is now the
   general form of both. */
.vz-dialog {
  box-sizing: border-box;
  width: min(34rem, calc(100vw - var(--vz-space-5)));
  padding: var(--vz-space-5);
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-lg);
  background: var(--vz-surface);
  color: var(--vz-ink);
  box-shadow: var(--vz-shadow-md);
}

.vz-dialog::backdrop { background: rgb(15 23 42 / 55%); }

.vz-dialog-head {
  display: flex;
  align-items: flex-start;
  justify-content: space-between;
  gap: var(--vz-space-3);
  margin-bottom: var(--vz-space-4);
}

.vz-dialog-title { margin: 0; font-size: var(--vz-fs-lg); color: var(--vz-ink); }

.vz-dialog-close {
  min-width: var(--vz-tap-min);
  min-height: var(--vz-tap-min);
  border: 0;
  border-radius: var(--vz-radius-md);
  background: none;
  color: var(--vz-muted);
  font-size: var(--vz-fs-lg);
  line-height: 1;
  cursor: pointer;
}

.vz-dialog-close:hover { background: var(--vz-surface-2); color: var(--vz-ink); }
.vz-dialog-close:focus-visible { outline: 2px solid var(--vz-primary); outline-offset: 2px; }

/* On a phone the dialog becomes a full-height sheet — `UI_CONVENTIONS §3`:
   "large forms may become full-screen sheets on mobile". */
@media (max-width: 44rem) {
  .vz-dialog {
    width: 100vw;
    max-width: 100vw;
    height: 100dvh;
    max-height: 100dvh;
    margin: 0;
    border: 0;
    border-radius: 0;
  }
}


/* The name cell is a link into the profile — the primary action on every row.
   It carries the underline only on hover so a dense table does not read as a
   wall of underlines, but it stays a real 24px target (WCAG 2.5.8). */
/* ⚠ 44px SINCE E25, AND THE MEASUREMENT IS WHAT UNBLOCKED IT. `OPERATING_MODE §7` item
   20 recorded this link at 25px against a `--vz-tap-min` of 44 and said what was
   missing: *"a row that has measured the consequence on every screen that would
   change"*. E25 captured all sixteen screens from the real controllers and measured it
   — on the members list at 375, with eight rows:

       the link   25px -> 44px
       a card row 135px -> 155px   (+20px)
       the page   1,962px -> 2,116px   (+154px, about 8%)

   ⚠ EIGHT PER CENT MORE SCROLLING IS THE PRICE, AND IT IS WORTH PAYING. This is the
   primary way into a record on a phone — a collector taps it all day — and it was the
   one control on the screen below the floor the product sets for itself. D11c set the
   precedent for fixing one of these in place. */
.vz-c-name a {
  display: inline-flex;
  align-items: center;
  min-height: var(--vz-tap-min);
  color: inherit;
  text-decoration: none;
}

.vz-c-name a:hover { text-decoration: underline; }

.vz-c-name a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
  text-decoration: underline;
}

/* ---------------------------------------------------------------------------
   🔴 DATA-CELL LINKS ARE NOT UNDERLINED AT REST (owner, 2026-09-24)

   `.vz-c-name a` above is the convention: no underline at rest, underline on hover and
   on focus. ⚠ SEVEN OTHER CELL TYPES CARRIED THE HOVER HALF AND NOT THE REST HALF, so
   their links were underlined all the time — and their `:hover { text-decoration:
   underline }` rules changed NOTHING, which is what proves the omission rather than a
   preference. An eighth, `.vz-c-invoice-no`, had neither half.

   ▶ Found by scanning for `X a:hover { underline }` with no matching rest-state `none`:
   seven dead rules against four live ones. The owner reported the eighth from a
   screenshot of the invoice register.

   ⛔ ONE BLOCK RATHER THAN EIGHT EDITS, so the convention has somewhere to live. A ninth
   cell type joins this selector list instead of re-deriving the rule and forgetting half
   of it — which is how all eight got here.

   ⚠ `:hover` AND `:focus-visible` STILL WIN, because `(0,2,1)` beats this rule's
   `(0,1,1)`. Nothing below needs reordering and nothing above is overruled.
   --------------------------------------------------------------------------- */
.vz-c-invoice-no a,
.vz-c-code a,
/* ⚠ `.vz-c-ref` IS SOMETIMES THE ANCHOR ITSELF AND SOMETIMES ITS PARENT, and the first
   draft of this list covered only the second — so the ledger reference stayed underlined
   while everything around it stopped. Both shapes, as the rules below it already pair. */
.vz-c-ref:is(a),
.vz-c-ref a,
.vz-kv dd a,
.vz-sub-meta a,
.vz-sub-value a,
.vz-table--report td a,
.vz-table--settlement-history .vz-c-when a {
  text-decoration: none;
}

/* ⚠ THE INVOICE SERIAL HAD NO HOVER RULE EITHER, so it gets the other half here — a link
   with no rest underline and no hover feedback is a control with no affordance at all,
   which is a worse screen than the underlined one the owner asked about. */
.vz-c-invoice-no a:hover { text-decoration: underline; }

.vz-c-invoice-no a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
  text-decoration: underline;
}

/* ===========================================================================
   Member sub-records: contacts and nominees (B2b)
   =========================================================================== */

.vz-panel-head {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  justify-content: space-between;
  gap: var(--vz-space-3);
  margin-bottom: var(--vz-space-4);
}

/* The head owns the spacing when it is present. */
.vz-panel-head .vz-panel-title { margin: 0; }

/* --- A list of sub-records ------------------------------------------------ */

.vz-sub-list {
  list-style: none;
  margin: 0;
  padding: 0;
}

/* 🐛🔴 F15 — THE PANEL SYSTEM HAD NO VERTICAL RHYTHM, AND FOUR SEAMS FELL THROUGH IT.
   Every gap in a panel is applied per-case, and four cases were never given one. All four
   measured at **0px** on 2026-09-23:

     ⓵ `.vz-sub-list` li → li          — /admin/support-sessions, FIVE distinct rules about
                                          break-glass access fused into one wall of text
     ⓶ stacked `.vz-sub-meta` spans    — /admin/users, the protected-root explanation
     ⓷ `.vz-kv` → `.vz-panel-note`     — /admin/audit, /admin/messaging
     ⓸ `.vz-panel` → `.vz-notice`      — /admin/settings, two CARDS with borders touching

   ⚠ THE CONTAINERS SHARE NOTHING BUT THE SYMPTOM. `.vz-kv` is `margin: 0` (38 views),
   `.vz-sub-list` is `margin: 0` (13 views across all three portals), `.vz-sub-meta` is a
   span made `display: block` by context, and `.vz-notice` is a sibling card. ▶ So this is
   four rules rather than one — a blanket `.vz-panel > * + *` would have reached dozens of
   pairs that are correctly tight, including `H1 → .vz-page-sub`, which measures 4px on
   purpose and is right.

   ⛔ AND THE STYLESHEET ALREADY CAUGHT THIS CLASS ONCE AND FIXED ONLY ITS OWN SEAM: the
   comment above `.vz-panel-subtitle:first-child` describes `.vz-sub-list`'s `margin-bottom: 0`
   collapsing a gap *"to exactly 0px so the heading sat on the list"*. The item-to-item gap
   was never looked at.

   ⚠ 12px, NOT 8: these items are multi-sentence paragraphs on the worst screen and one-liners
   on the mildest (`/admin/settings`'s ignored rows). 12px reads as separation at paragraph
   length without making a short list look sparse — checked on both. */
.vz-sub-list li + li {
  margin-top: var(--vz-space-3);
}

.vz-sub-meta + .vz-sub-meta {
  margin-top: var(--vz-space-2);
}

/* 🐛 AND THE MEMBER PORTAL IS THE CASE THE RULE ABOVE WAS NOT WRITTEN FOR (owner,
   2026-09-28, point 3: *"Too much spacing in members Auction results page"*).

   ⚠ **`F15` CHOSE 8px FOR `/admin/users`' PROTECTED-ROOT EXPLANATION — MULTI-SENTENCE
   PARAGRAPHS.** A member's auction card stacks three short facts about **one** draw
   — *"Installment 1 · Group KCF-001 · ticket 1"*, *"Held 28 Sep 2026"*, *"Won by: …"* —
   and at 8px they read as three separate paragraphs instead of one block of detail.
   MEASURED at 10px each (8px margin on top of the column's own 2px gap) against the
   trailing column's 2px on the same row: **the two halves of one card had different
   rhythms**, and the left one was the loose one.

   ▶ **SCOPED TO `.vz-portal`, AND EVERY SCREEN INSIDE IT WAS OPENED.** Stacked
   `.vz-sub-meta` occurs on exactly two: `/p/…/auctions` and `/p/…/notifications`, and on
   both the lines are short facts about one thing. `/p/…/profile` has none and the home
   screen never stacks two. ⛔ The staff portals keep `F15`'s 8px untouched. */
.vz-portal .vz-sub-meta + .vz-sub-meta {
  margin-top: 0;
}

/* ⚠ A FIFTH SEAM, FOUND BY LOOKING AT THE FIX FOR THE FIRST FOUR. With the list items
   separated, `/admin/support-sessions`' closing note was suddenly the only thing still
   welded to what preceded it — `.vz-sub-list` ends, `.vz-panel-note` begins, gap 0.
   ▶ Fixing a rhythm exposes the next place it is missing, which is an argument for looking
   again AFTER a layout change rather than only before one.

   🐛 AND A SIXTH, F27 (2026-09-24), FOUND THE SAME WAY THE FIFTH WAS. The billing panel on
   `/admin/organizations/{id}` ends `<details>` (a collapsed form) then `.vz-panel-note` (what
   the panel cannot change), and that pair measured **0px** against 8, 16 and 20 everywhere
   else in the same panel. ⚠ IT IS A NEW PAIR RATHER THAN A MISSED ONE: no screen in the
   product had ever put a note directly after a `<details>` until two forms moved onto a page
   that also explains itself. ▶ `<details>` has no margin of its own, so the note's `margin: 0`
   welds the sentence to the disclosure triangle above it and it reads as part of the form.

   ⚠ SCOPED TO `.vz-panel-note` AND NOT TO `details + *`, which is the fifth seam's rule and
   the reason the first four were four rules: this pair is a paragraph following a control, and
   a blanket sibling selector would reach the `<details>`-to-`<details>` and
   `<details>`-to-heading pairs that already have a gap and would then have two.

   ⏹ **`.vz-kv + .vz-panel-note` WAS THE THIRD SELECTOR HERE AND HAS MOVED** — to the rule that
   answers *"what does a `.vz-kv`'s neighbour get"*, below the pager one. ▶ **It was this
   rule's only `.vz-kv` clause and it was a patch for one neighbour class**, which is exactly
   the shape `.vz-panel + .vz-notice` was folded in to stop. `§7` item 107: amended where it
   stands, because the premise is the part that moved. */
.vz-sub-list + .vz-panel-note,
details + .vz-panel-note {
  margin-top: var(--vz-space-3);
}


.vz-sub-row {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  justify-content: space-between;
  gap: var(--vz-space-2) var(--vz-space-3);
  padding: var(--vz-space-3) 0;
  border-bottom: 1px solid var(--vz-border);
}

.vz-sub-row:last-child { border-bottom: 0; }
.vz-sub-row:first-child { padding-top: 0; }

.vz-sub-main {
  display: flex;
  flex-direction: column;
  gap: 2px;
  min-width: 0;
}

.vz-sub-value {
  color: var(--vz-ink);
  font-size: var(--vz-fs-sm);
  font-weight: var(--vz-fw-medium);
  /* Numbers are read digit by digit off this list. */
  font-variant-numeric: tabular-nums lining-nums;
  overflow-wrap: anywhere;
}

/* The document link on a member's file list (B7). It is the ONLY way to the file — the
   object key is unguessable and never rendered — so it has to be a real 24px target, the
   same rule `.vz-c-name a` and `.vz-c-code a` already carry. */
.vz-sub-value a {
  display: inline-flex;
  align-items: center;
  min-height: 24px;
  color: inherit;
}

.vz-sub-value a:hover { text-decoration: underline; }

.vz-sub-value a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* 🐛 A LINK IN A `.vz-sub-meta` MEASURED **19px TALL** (`§O1d`, at 375 on the member
   portal) — under WCAG 2.5.8's 24×24 minimum, on the control that is the ONLY route from
   a member's notification feed to their passbook. Six of them on one screen.

   ⚠ **24px AND NOT `--vz-tap-min`'s 44**, deliberately. `.vz-sub-value a` two rules up
   already carries exactly this floor with exactly this argument — *"it is the ONLY way to
   the file … so it has to be a real 24px target"* — and inventing a second number for the
   same component on one screen is how a design system acquires two answers to one question.
   ▶ 44px is the floor for a CONTROL (`.vz-tab` takes it — `.vz-portal-nav a` did until that
   component was deleted on 2026-09-28); this is a text link in a
   line of meta text, and `§7` item 20 records what raising one of those product-wide cost:
   +8%% page height on the members list, argued and accepted once.

   ~~⚠ It reaches two admin screens (`admin/jobs`, `admin/messaging`) and grows each link by
   5px. Measured rather than assumed: those are the only other two views in the product with
   an `<a>` inside a `.vz-sub-meta`.~~

   🔴 **`F73` — THAT MEASUREMENT WENT STALE, AND IT IS NOW FIVE ADMIN VIEWS, NOT TWO.**
   Re-measured 2026-09-26: `admin/audit/index`, `admin/messaging/template` and
   `admin/support-sessions/index` were all written after the sentence above and inherit this
   rule without ever being weighed against it. The reasoning is unaffected — the rule is the
   right default for a text link in a line of meta text — but a reader revisiting the
   24-versus-44 trade would have under-counted its reach by more than half.

   ⚠ **AND NOTHING ASSERTED IT**, which is why it could rot silently.
   `Tests\Unit\StylesheetReachTest` now pins the set: a sixth view joining it fails there,
   naming this comment, rather than quietly widening a decision's blast radius.

   ── ✅ AND THE GUARD FIRED, THE FIRST TIME THE SET MOVED (2026-09-28) ──
   `portal/home.php` **LEFT** this set. The owner asked for *"a BUTTON for 'See the passbook'
   rather than a link"* (point 5), and that control — the one route from a member's group to
   its passbook — was an `<a>` inside `.vz-sub-meta`, i.e. **capped at 24px against
   `--vz-tap-min`'s 44**, on a phone, as the primary action of the screen.

   ▶ **SO THE FIX IS A TAP-TARGET FIX AS WELL AS A PROMINENCE ONE.** As a `.vz-action` it
   takes the full 44px floor. ⚠ The cap itself is unchanged and still right for the six views
   that remain: a text link in a line of meta text is not a control. */
.vz-sub-meta a {
  display: inline-flex;
  align-items: center;
  min-height: 24px;
  color: inherit;
}

.vz-sub-meta a:hover { text-decoration: underline; }

.vz-sub-meta a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* 🔴 `ⓠ4` — THE MEMBER PORTAL IS TOUCH-ONLY, SO `:hover` IS NOT AN AFFORDANCE THERE.
   ---------------------------------------------------------------------------
   The rest-state convention above — no underline until hover — comes from `.vz-c-name a` and
   was reasoned for **operator table cells**, where a mouse exists. `.vz-portal` has no mouse:
   *"See the passbook"* is the only route from a member's home screen to their own money, and
   without this rule it renders in `color: inherit` from its muted parent, with no underline,
   **the same colour as the static "Group PG1 · ticket 1" on the line above it.**

   ⚠ **AN UNDERLINE AND NOTHING ELSE.** Not a colour, not a weight, not a box — and
   deliberately **not** the tap target. `.vz-sub-meta a`'s 24px is a considered number: WCAG
   2.5.8 (AA) is 24×24, `--vz-tap-min`'s 44 is 2.5.5 (AAA), and `§7` item 20 measured what
   raising these product-wide costs at **+8% page height on the members list**. An underline
   adds no height, so none of that argument is reopened here.

   ⛔ **SCOPED TO THE PORTAL, WHICH IS THE WHOLE POINT.** The operator app keeps the
   convention where its reasoning holds. `(0,2,1)` beats the `(0,1,1)` list above it, so
   nothing needs reordering. */
.vz-portal .vz-sub-meta a { text-decoration: underline; }

.vz-sub-meta {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-2);
  color: var(--vz-subtle);
  font-size: var(--vz-fs-xs);
}

/* The reason a surety was released — free text, up to 500 characters (D1c).

   ⚠ IT HAS ITS OWN CLASS BECAUSE IT BORROWED `.vz-c-moved` AND INHERITED A BUG.
   That rule is the schedule's "moved from" line, where `white-space: nowrap` is right
   because the content is a short date. A release reason is a sentence somebody types,
   and nowrap sent 737px of horizontal overflow through the whole page — found by
   measuring with a REALISTIC reason rather than the 45-character one that happened to
   be in the demo data, which fitted and measured zero. Same blind spot C2c recorded:
   content that fits is not the same as layout that holds. */
.vz-sub-reason {
  display: block;
  color: var(--vz-subtle);
  font-size: var(--vz-fs-xs);
  overflow-wrap: anywhere;
}

/* 🔴🔴 F41 — THIS ROW USED TO SCROLL A PHONE SIDEWAYS, AND `flex: 0 0 auto` WAS WHY.
   Raised 2026-09-25 off the 1024px render, and **measuring it found the worse half**: a
   surety row's three forms are **441px of content that never shrank at any width**, against a
   container of **286px at 375 and 231px at 320**. ▶ It burst its panel by 155px and 210px, and
   `/app/members/{id}` carried **118px of horizontal PAGE overflow at 375 and 173px at 320**.
   ⛔ Nothing measured it, because the row FITS at every width the design sweep looks at —
   `§7` item 63's blind spot exactly: content that fits is not a layout that holds.

   ▶ **TWO DECLARATIONS, BOTH LOAD-BEARING, AND A THIRD WAS DROPPED FOR DOING NOTHING.**
   Measured one at a time at 375/320 (page overflow in px):

       shipped `flex: 0 0 auto`        118 / 173
       `flex-wrap: wrap` alone         118 / 173   ← the row cannot shrink, so it never wraps
       `min-width: 0` alone            118 / 173
       `flex: 0 1 auto` alone           32 /  87   ← shrinks to 355, still overflows
       `flex: 0 1 auto` + wrap           0 /   0   ← the forms stack, and it fits
       all three                         0 /   0   ← `min-width: 0` adds NOTHING, so it is not here

   ⚠ **SHRINKING DOES NOT SQUEEZE THE CONTROLS**, which is what `0 0 auto` was guarding: the
   forms wrap onto their own lines instead. ▶ **Re-measured on the shipped file at 1440 · 1024 ·
   768 · 600 · 480 · 375 · 320: ZERO page overflow at every one of them**, the reason field and
   its button together down to 375, and the button wrapping beneath it only at 320.
   The contact row (142px) and `/app/users/{id}` were never affected either way. */
.vz-sub-actions {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-1);
  flex: 0 1 auto;
}

/* 🐛 A BUTTON 2px BELOW THE LINE OF TEXT ABOVE IT (owner, 2026-09-28, point 4:
   *"Add space a little bit above see the passbook button"*). MEASURED on the member portal's
   home screen: `.vz-sub-main` is a stacked column with `gap: 2px` — right for two lines of
   text, and **the passbook button is a third child of that column**, so it inherited a 2px
   separation from the meta line and read as another line rather than as a control.

   ⚠ **SCOPED TO `> .vz-sub-actions` INSIDE A `.vz-sub-main`, WHICH IS ONE CALLER TODAY.**
   The other two views that use `.vz-sub-actions` — `/app/members/{id}` and `/app/users/{id}`
   — put it **beside** `.vz-sub-main` as a sibling, where the row's own `gap` already spaces
   it. A rule on `.vz-sub-actions` alone would have moved those too, and the drive fixture
   draws neither of the states they carry.

   ▶ **12px RATHER THAN 8, AND THE ROW ITSELF IS THE ARGUMENT**: `.vz-sub-row` is
   `padding: var(--vz-space-3) 0`, so the button's distance from its own text now matches the
   distance between one seat and the next. Both were drawn and compared. */
.vz-sub-main > .vz-sub-actions { margin-top: var(--vz-space-3); }

/* 🔴 "Nothing due" IS A STATE, AND EVERY OTHER STATE IN THIS PRODUCT IS A PILL (owner,
   2026-09-28, point 5). It rendered as `.vz-sub-value` — the same weight as the money figure
   on the row beneath it — so the one row a member is *not* worried about looked like the one
   they are.

   ⚠ `align-self` IS THE WHOLE RULE, BECAUSE A FLEX ITEM'S `inline-flex` IS BLOCKIFIED. Inside
   a `column` container with the default `stretch`, the pill would have spanned the column's
   full width instead of hugging its label — `.vz-table--receipts .vz-c-status .vz-badge` two
   thousand lines down solves the same thing with `width: max-content`.

   🔴🔴 **`flex-start`, AND THE FIRST DRAFT OF THIS RULE SAID `flex-end` AND WAS A DEFECT.**
   The reasoning was that this pill lives in the row's TRAILING column, so the trailing edge
   is the one it reads from. ⛔ **`/p/…/auctions` puts a `.vz-badge` in the LEADING column** —
   *"You won this one"* — and `flex-end` pushed it **70px in from the left margin**, adrift
   from every other line on the card. ▶ Found by opening the next screen, not by any check
   here: `a-fix-that-stops-one-element-short`, and the same rule's own blast radius.

   ⚠ **THE TRAILING COLUMN LOSES NOTHING BY IT.** `.vz-sub-row` is `space-between` and the
   column is `flex: 0 1 auto`, so it shrink-wraps to the pill and its right edge is the row's
   right edge whichever way the pill aligns inside it — measured at 816px both ways. */
.vz-sub-main > .vz-badge { align-self: flex-start; }

/* ⚠ AND A PILL THAT FOLLOWS TEXT IN THAT COLUMN NEEDS AIR THE TEXT DOES NOT. With the
   portal's stacked metas tightened to 2px above, `/p/…/auctions`' *"You won this one"* sat
   2px under *"Held 28 Sep 2026"* — a bordered control welded to a line of prose. Text lines
   at 2px read as one block, which is the point; a pill at 2px reads as a mistake.

   ▶ **`* +` MEANS IT NEVER FIRES ON A PILL THAT LEADS ITS COLUMN**, which is how
   `/p/…/notifications` draws *"Draw held"* — first child, nothing above it to be spaced from.
   ⚠ **AND IT IS `.vz-portal`-SCOPED BECAUSE `/app/members/{id}` PUTS SEVEN BADGES IN THIS
   COLUMN**, most of them inline inside a `.vz-sub-meta` rather than a child of the column,
   and the drive fixture draws almost none of those states — `attribute-before-building-the-
   guard`: a rule whose blast radius I cannot look at is a rule I do not ship. */
.vz-portal .vz-sub-main > * + .vz-badge { margin-top: var(--vz-space-2); }

/* --- The nominee set ------------------------------------------------------ */

.vz-nominee-head {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  justify-content: space-between;
  gap: var(--vz-space-3);
  margin-bottom: var(--vz-space-3);
}

.vz-nominee-head .vz-panel-title { margin: 0; }

.vz-nominee-remove {
  display: inline-flex;
  align-items: center;
  gap: var(--vz-space-2);
  min-height: var(--vz-tap-min);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
  cursor: pointer;
}

/* ⚠ THE RUNNING TOTAL IS RENDERED BY THE SERVER, not computed in JavaScript, so the
   number shown is the number that was validated. Out of balance it gains a border and
   a weight — never colour alone (WCAG 1.4.1), and the message beside it says the
   same thing in words. */
.vz-nominee-total {
  margin-top: var(--vz-space-4);
  padding: var(--vz-space-3) var(--vz-space-4);
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface-2);
  color: var(--vz-ink);
  font-size: var(--vz-fs-sm);
  font-variant-numeric: tabular-nums lining-nums;
  text-align: right;
}

.vz-nominee-total.is-unbalanced {
  border-color: var(--vz-warning);
  border-width: 2px;
  font-weight: var(--vz-fw-bold);
}


/* ===========================================================================
   THE CHIT-GROUP WIZARD AND LIFECYCLE (B3)
   ---------------------------------------------------------------------------
   Same rules as everything above: no colour is introduced, every value is a
   `--vz-*` token, and nothing here needs JavaScript. The wizard's stepper is a
   list of links to real URLs and the lifecycle actions are POST forms, so this
   file styles navigation and text — it never becomes the thing that makes the
   screen work.
   =========================================================================== */

/* --- The stepper ---------------------------------------------------------- */

/* ⚠ AN ORDERED LIST, because the steps genuinely have an order and a screen
   reader should be able to say "3 of 5". A row of styled divs announces nothing.
   It scrolls horizontally on a narrow screen rather than wrapping into an
   unreadable stack of five full-width rows — and the scroll container is the
   nav, so the PAGE never scrolls sideways. */
.vz-steps {
  margin-bottom: var(--vz-space-4);
  overflow-x: auto;
  -webkit-overflow-scrolling: touch;
}

.vz-steps-list {
  display: flex;
  gap: var(--vz-space-2);
  margin: 0;
  padding: 0 0 var(--vz-space-2);
  list-style: none;
  min-width: min-content;
}

/* ⚠ `position: relative` IS THE OTHER HALF OF THE OVERFLOW FIX. A frozen step's
   visually-hidden note is `position: absolute`, and with no positioned ancestor it
   resolves against the INITIAL containing block — so its static position, deep
   inside a stepper scrolled 800px wide, extended the document even at 1px wide.
   Making the `<li>` the containing block keeps it inside the scroll container
   where it belongs. Measured: 375px viewport, document scrollWidth 502 → 375. */
.vz-step { flex: 0 0 auto; position: relative; }

/* ⚠ `:not(.vz-sr-only)` IS LOad-BEARING, NOT TIDINESS. A frozen step carries a
   visually-hidden "(cannot be changed at this stage)" as a direct child of the
   `<li>`, so a bare `.vz-step > span` also styled THAT — giving an
   absolutely-positioned 1px node padding and a border, which pushed the document's
   scrollWidth 127px past the viewport at 375px. The page scrolled sideways and
   nothing visible was out of place, which is why it was found by measuring rather
   than by looking. */
.vz-step > a,
.vz-step > span:not(.vz-sr-only) {
  display: flex;
  align-items: center;
  gap: var(--vz-space-2);
  /* The 44px tap target `BRANDING` pins, met by the control itself rather than
     by a margin that only looks like room. */
  min-height: var(--vz-tap-min);
  padding: var(--vz-space-2) var(--vz-space-3);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
  text-decoration: none;
  white-space: nowrap;
}

.vz-step > a:hover {
  border-color: var(--vz-border-strong);
  color: var(--vz-ink);
}

.vz-step > a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* The current step. ⚠ Weight and border as well as colour — `aria-current` tells
   assistive technology, and these two tell anyone who cannot distinguish the
   green (WCAG 1.4.1). */
.vz-step.is-current > span:not(.vz-sr-only) {
  border-color: var(--vz-primary-strong);
  border-width: 2px;
  color: var(--vz-ink);
  font-weight: var(--vz-fw-medium);
}

/* A step frozen at this status. Dashed, muted, and carrying a visually-hidden
   "(cannot be changed at this stage)" in the markup — the pattern is not the
   only signal. */
.vz-step.is-frozen > a,
.vz-step.is-frozen > span:not(.vz-sr-only) {
  border-style: dashed;
  color: var(--vz-subtle);
}

.vz-step-no {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  width: 1.5rem;
  height: 1.5rem;
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface-2);
  font-size: var(--vz-fs-xs);
  font-variant-numeric: tabular-nums lining-nums;
}

/* ⚠ THE BUTTON FILL, NOT THE BRAND-TEXT TOKEN (E15). This is the wizard's current-step
   badge: a filled circle carrying `--vz-primary-contrast` on top, which makes it the same
   KIND of thing as the primary button and subject to the same measurement.
   `--vz-primary-strong` is the token for text ON a surface, and it happened to be legible
   here only because both tokens used to be the same colour. */
.vz-step.is-current .vz-step-no {
  background: var(--vz-primary-button);
  color: var(--vz-primary-contrast);
}

/* --- A frozen field ------------------------------------------------------- */

/* ⚠ TEXT, NOT A DISABLED INPUT. A disabled control is skipped by keyboard
   navigation, so the value becomes unreadable to exactly the users who most need
   it read out — and it still looks like a form, which invites staff to try. This
   says what is true: the value stands, and this is not where it changes. */
.vz-field-static {
  margin: 0;
  padding: var(--vz-space-2) 0;
  min-height: var(--vz-tap-min);
  display: flex;
  align-items: center;
  color: var(--vz-ink);
  font-size: var(--vz-fs-base);
  font-variant-numeric: tabular-nums lining-nums;
  border-bottom: 1px dashed var(--vz-border);
}

/* --- Lifecycle transitions ------------------------------------------------ */

.vz-transition {
  padding: var(--vz-space-4) 0;
  border-top: 1px solid var(--vz-border);
}

.vz-transition:first-of-type {
  padding-top: 0;
  border-top: 0;
}

.vz-transition-title {
  margin: 0 0 var(--vz-space-1);
  font-size: var(--vz-fs-base);
  font-weight: var(--vz-fw-medium);
  color: var(--vz-ink);
}

.vz-transition-note {
  margin: 0 0 var(--vz-space-3);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
  line-height: var(--vz-lh-base);
}

.vz-transition textarea.vz-control {
  min-height: 4rem;
  resize: vertical;
  font-family: inherit;
}

/* ⚠ WHAT IS IN THE WAY, NOT A DISABLED BUTTON. Each item names a field and links
   at the step that holds it. The marker is a rule rather than a bullet glyph, so
   the list reads as an obstruction rather than as a checklist of things done. */
.vz-blockers {
  margin: 0;
  padding: 0;
  list-style: none;
  display: flex;
  flex-direction: column;
  gap: var(--vz-space-2);
}

.vz-blockers li {
  padding-left: var(--vz-space-3);
  border-left: 3px solid var(--vz-warning);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
  line-height: var(--vz-lh-base);
}

.vz-blockers a {
  display: inline-flex;
  align-items: center;
  min-height: 24px;
  margin-left: var(--vz-space-2);
  color: var(--vz-primary-strong);
  white-space: nowrap;
}

.vz-blockers a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* --- List columns specific to groups -------------------------------------- */

/* Money and counts get tabular figures and right alignment: a column of rupees
   that does not line up on the decimal point cannot be scanned, and scanning it
   is the only reason it is on the screen. */
/* 🔴 THE SUBSCRIPTION REGISTER'S PERIOD CELL WRAPS, WHERE `.vz-c-when` OTHERWISE DOES NOT
   (`§Q-R4`). "01 Aug 2026 to 31 Aug 2026" is 26 characters the base `nowrap` will not break,
   and measured it floored the column at 196px — the widest of five — putting the table 734px
   inside a 740px wrap in the narrow band just above its card breakpoint.

   ⚠ AND IT CARRIES A `min-width` WITH THE WRAP, WHICH IS ROW 14's PAIRED RULE. Wrapping alone
   makes the column's min-content the longest TOKEN ("2026"), and the table algorithm is then
   free to crush it into a vertical stack of three-character lines at some width in the middle
   of the range — at zero page overflow, because nothing is overflowing. The floor is what
   keeps "01 Aug 2026" on one line while allowing the break before "to".

   ⚠ A COLUMN OF TWO DATES IS NOT PROSE, so this is not the `.vz-c-moved` trap in reverse:
   there is exactly one place it may break and it is between the two halves of the phrase. */
.vz-table--subscription .vz-c-when {
  white-space: normal;
  min-width: 7rem;
}

/* ⚠ `.vz-c-paid` SHARES EVERY DECLARATION AND IS A SEPARATE SELECTOR ON PURPOSE (`§Q-R4`).
   It is a second money column on the same row as `.vz-c-money` — the total and what has
   been received against it — and the phone card has to place the two in different areas.
   A single class on both cells would make that impossible, and `grid-area` is the whole
   reason the card rules exist. ⛔ Do not "tidy" this into one selector. */
/* ⚠ F38 TOOK `text-align: right` OUT OF HERE — it was `0,1,0` and never applied. Every
   figure column's alignment is one block in the base table section now; this rule keeps the
   declarations that actually worked. */
.vz-c-money,
.vz-c-paid {
  white-space: nowrap;
  font-variant-numeric: tabular-nums lining-nums;
}

.vz-c-filled,
.vz-c-term,
.vz-c-next {
  white-space: nowrap;
  font-variant-numeric: tabular-nums lining-nums;
  color: var(--vz-muted);
}

.vz-c-unset { color: var(--vz-subtle); }

/* ⚠ THE GROUP TABLE NEEDS ITS OWN CARD LAYOUT, AND THAT IS WHY `.vz-table--chits`
   EXISTS. The phone rules above turn any `.vz-table` row into a card using
   `grid-template-areas` that name the MEMBER columns — a chit row's cells would
   match none of those names, get auto-placed into implicit rows, and come out in
   an order nobody chose. The areas are therefore re-declared for this table.

   Same principle as the member list, which is the point: nothing is hidden at
   375px. Staff on a phone are the ones being asked "is that group full yet", and
   a column that vanishes is the answer they came for. */
@media (max-width: 64rem) {
  /* ⚠ `actions` IS A NAMED AREA HERE FOR THE SAME REASON EVERY OTHER CELL IS (E6b).
     An unnamed cell is auto-placed in an order nobody chose and still measures zero
     overflow, because it fits — the blind spot this modifier exists to avoid. */
  .vz-table--chits tr {
    grid-template-areas:
      'name    status'
      'code    value'
      'term    filled'
      'next    next'
      'actions actions';
  }

  .vz-c-money  { grid-area: value; justify-self: end; }
  .vz-c-term   { grid-area: term; }
  .vz-c-filled { grid-area: filled; justify-self: end; }
  .vz-c-next   { grid-area: next; font-size: var(--vz-fs-xs); }

  .vz-c-money,
  .vz-c-filled,
  .vz-c-term,
  .vz-c-next { text-align: left; }

  /* 🔴 F38: THE CARD KEEPS ITS LEFT EDGE, AND THIS RULE IS WHY IT STILL CAN.
     Turning the desktop alignment on meant `.vz-table .vz-c-money` at `0,2,0`, which
     **outranks the `0,1,0` rule directly above** — harmless for `.vz-c-money` itself,
     because the generic placement two lines up shrink-wraps it with `justify-self: end`
     and the alignment inside a box exactly as wide as its text is invisible.
     ⛔ `.vz-c-total` AND `.vz-c-penalty` HAVE NO SUCH PLACEMENT. Measured 2026-09-24: five
     card cells across `--dues`, `--collect` and `--arrears` fill their whole grid area with
     no card opinion at all, so the desktop fix would have swung them to the right edge of a
     phone card as a side effect.
     ⚠ **THE CARD HAS NO HEADER ROW, SO IT IS NOT WHAT F38 IS ABOUT** — `thead` is
     screen-reader-only below 64rem. ▶ Changing a phone layout nobody asked about, on
     operator screens this row's fixture cannot reach, is not a thing to do by accident.
     `0,2,0` and later in the file, so it wins on source order. */
  .vz-table .vz-c-total,
  .vz-table .vz-c-penalty { text-align: left; }

}

/* 🔴 SPLIT OUT WHEN THE CARD BREAKPOINT MOVED TO 64rem (item 63's row), for the reason the
   toolbar rules carry: a stepper and a blocker link are PHONE layout and say nothing about
   whether a table is a card. ⚠ `.vz-steps-list`'s own comment names 375px as the width it is
   about, which is the tell — a rule whose comment names a phone width belongs at the phone
   breakpoint. */
@media (max-width: 44rem) {
  /* A five-pill stepper does not fit a 375px screen, and wrapping it into five
     full-width rows pushes the form itself below the fold. It scrolls inside its
     own container instead — the page never scrolls sideways. */
  .vz-steps-list { padding-bottom: var(--vz-space-3); }

  .vz-blockers a {
    display: flex;
    margin-left: 0;
    margin-top: var(--vz-space-1);
    min-height: var(--vz-tap-min);
    white-space: normal;
  }
}

/* A link inside a page subtitle — the group name on a schedule or reschedule
   screen (B4). It sits in a sentence, so WCAG 2.5.8's inline-text exemption
   arguably covers it; the project's own standard is "no target under 24px,
   measured", and B1's `.vz-facts dd a` note already refused to lean on that
   exemption once. Cheaper to meet the bar than to argue about it. */
.vz-page-sub a {
  display: inline-flex;
  align-items: center;
  min-height: 24px;
  color: inherit;
}

.vz-page-sub a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* --- The installment schedule (B4) ---------------------------------------- */

.vz-c-no {
  width: 3rem;
  font-variant-numeric: tabular-nums lining-nums;
  color: var(--vz-subtle);
}

.vz-c-when { white-space: nowrap; }

/* Where a moved installment came from, and when it was actually held. Smaller and
   muted, on its own line: it is context for the date above it, not a second date
   competing with it. */
.vz-c-moved {
  display: block;
  color: var(--vz-subtle);
  font-size: var(--vz-fs-xs);
  white-space: nowrap;
}

/* 🔴 EXCEPT INSIDE A NOTE CELL, WHERE IT IS A SENTENCE RATHER THAN A DATE OR A CODE
   (item 63's row). `.vz-c-moved` is `nowrap` for its original subject — a moved installment's
   origin date, and the member code under a name — and that is right: both are dictated. But
   `/app/collections/followups` puts *"No messaging consent"* in one, and a sentence in a
   `nowrap` span sets the whole COLUMN's minimum to its entire line: **121px of an unbreakable
   phrase inside a 153px column**, on the widest register in the product.

   ▶ **THE PAIRING RULE, WHICH IS `§Q10`'s PRECEDENT EXACTLY.** That row fixed the same shape by
   scoping `overflow-wrap: anywhere` to `.vz-table--config .vz-c-name .vz-mono` and said why:
   never on `.vz-mono` itself, *"which is worn by member codes and phone numbers that must NOT
   break"*. ⛔ Same here — this is scoped to `.vz-c-note`, never to `.vz-c-moved`. **One class,
   two jobs, and only the prose one relaxes.** */
.vz-c-note .vz-c-moved { white-space: normal; }

/* ⚠ `width: 1%` IS THE SHRINK-TO-FIT IDIOM, AND IT IS WHAT STOPS THE GLYPHS FLOATING
   (E6b). `.vz-table` is `width: 100%`, so a table with room to spare hands the slack to
   its columns — and the ACTIONS column, being last, showed it all as a gap around the
   icons. The owner reported it as *"too much spacing between the icon and the actions
   header"*.
   🔴 F39 CORRECTED THIS SENTENCE: it used to say *"being last and right-aligned"*, and the
   column has never been right-aligned in a browser — see the rule below. **`width: 1%` is
   the whole of that fix**, which is why removing the dead alignment costs nothing here.

   A percentage this small cannot force a column below its own min-content width, so a
   cell that genuinely needs room still gets it; what changes is that the SLACK goes to
   the data columns instead. Measured on every list rather than assumed — the roster and
   the two collection screens hold wide action cells and are checked in `§E6b`. */
/* 🔴 THE ACTION CELL WRAPS SINCE ITEM 63's ROW, AND `§8a` ROW 7i IS WHY.
   ⏸ It carried `white-space: nowrap`, and 7i recorded the consequence as a DEFECT rather than
   as the intent: *"`.vz-c-actions` is `white-space: nowrap` and `.vz-c-actions form` is
   `display: inline-flex`, so **two `<form>`s in one action cell sit on one line that can never
   break**. Any table whose action cell holds a whole form is one control away from overflowing
   its wrap."* ▶ **Relaxing it is that finding's own remedy**, not a change against it: the
   controls stack instead of forcing every data column narrower.

   Measured at 1025 with nothing else changed: `.vz-table--chits` **37px → 0** and the members
   register **83px → 0**, both closed outright, with **no effect on any other register** — so
   this bought two of item 63's remaining five and cost nothing. Proved both ways.

   ⚠ `width: 1%` IS UNTOUCHED. The shrink-to-fit idiom is what gives the slack to the data
   columns and is argued above; `§Q-R5` found it was the real cause of a clipped control on one
   screen and fixed that by opting THAT table out, which is still the right shape for a cell
   holding a whole form. **What changed in item 63's row is only whether the cell may use a
   second line.** */

/* ═══ 🔴🔴 F39 — `text-align: right` IS GONE FROM HERE, AND IT WAS NEVER ON THE PAGE ═══════
   Filed by F38's measurement 2026-09-24, answered by the owner 2026-09-25: **delete the
   declaration.** It sat here unprefixed at `0,1,0`, and `.vz-table th, .vz-table td` is
   `0,1,1` — so all 41 action cells in the product have always rendered LEFT.

   🔴 **AND THE REASON IT IS DELETED RATHER THAN MOVED IS MEASURED, NOT PREFERRED.** Turning it
   on the way F38 did — a `.vz-table`-prefixed rule at `0,2,0` — was rendered and measured
   across twelve registers first: **the buttons move 0px on six of them and at most 11px on the
   other six**, because `width: 1%` has already shrink-wrapped the column to its own content.
   ▶ **The alignment has nothing left to align.** Paying for that would also mean putting
   `.vz-c-actions` on **37 `<th>` headers** — otherwise a left "ACTIONS" sits over right-aligned
   buttons, which is the exact defect F38 was raised to fix.

   ⚠ **THE SIXTEEN CARD RULES THAT SAY `text-align: left` ARE NOW REDUNDANT, AND THEY STAY.**
   Below 64rem `.vz-c-actions` is a grid area rather than a cell, and each of those rules is
   doing three other things beside the alignment. **They are the strongest evidence the
   right-align was once intended** — F38's own argument — and the owner's answer is that intent
   never reached a screen and is not worth 37 edits to revive. ⛔ Removing them is a separate
   sweep, not this row's. */
.vz-c-actions { width: 1%; white-space: normal; }

/* 🐛 ⚠ AND FOUR TABLES OPT OUT, BECAUSE THEIR ACTION CELLS CARRY MORE THAN GLYPHS.
   Measured, not guessed: with `width: 1%` on everything, the roster's cell came out
   **35px wide and 165px tall** — six actions stacked one per line, because a percentage
   that small lets a wrapping flex row collapse to its narrowest item. The two collection
   screens and the receipts table hold a whole form each, with a reason field and a
   step-up password inside the cell.

   ⚠ ENUMERATED BY MODIFIER RATHER THAN BY `:has()`. A `:has()` guard would be tidier and
   it would be LOAD-BEARING — a browser without it squashes a money form's password field
   to 35px. `.vz-panel-title:has(+ .vz-panel-note)` is an enhancement and may fail softly;
   this cannot, so it is a list, and a new table with a wide action cell joins it. */
/* ⚠ THE ROSTER LEFT THIS LIST IN E6e. It was here because its action cell was a wrapping
   flex box that `width: 1%` could collapse; the flex is on an inner `<div>` now, so the
   cell shrinks to its glyphs like every other one. **The rule for membership is that the
   cell holds a whole FORM** — a reason field and a step-up password, or a file input —
   rather than a row of glyphs. ⚠ THE COUNT IS DELIBERATELY NOT STATED: an earlier draft
   said *"the four that remain"* and was accurate until a fifth joined, which is `§6`'s
   *"never put a count in a heading; the list counts itself"* arriving in a stylesheet
   comment.

   🐛 `.vz-table--onboarding` JOINED 2026-09-15 AND HAD BEEN MISSING SINCE M3. Its action
   cell holds an upload form — a native `<input type="file">` beside its button — and with
   `width: 1%` the column came out **135px** at 1049, which squeezed the input to **26px
   against the 219px it needs to draw its own "Choose file" text**. That is the clipped
   control in the owner's screenshot, and the mechanism was written down here two rows
   before the table that needed it existed. */
.vz-table--dues .vz-c-actions,
.vz-table--receipts .vz-c-actions,
.vz-table--collect .vz-c-actions,
.vz-table--runs .vz-c-actions,
.vz-table--onboarding .vz-c-actions { width: auto; }

/* 🐛 ⚠ AND THOSE SAME FOUR MUST BE ALLOWED TO WRAP — MEASURED, `§8a` ROW 7i.
   `.vz-c-actions` is `white-space: nowrap` and `.vz-c-actions form` is `inline-flex`, so
   TWO forms in one cell sit on one line that can never break. The seat's payment history
   has had two since C2b — a pending cheque shows *Mark cleared* beside the whole
   *Why / password / Mark bounced* form — and it ran the receipts table to **1038px inside
   a 958px wrap at 1440**, with the retraction button 79px off the visible edge. Row 7i
   found it with its own control capture while adding a THIRD form, which would have taken
   it to 1089.

   🔴 THE `nowrap` ABOVE IS RIGHT AND THIS IS NOT A REVERSAL OF IT. Its argument
   (`.vz-row-actions`, `§E6c`) is that a wrapping flex row of GLYPHS has a min-content
   width of one glyph, so the table algorithm squashes the column and stacks the icons.
   These four cells hold no glyph row: they hold whole forms with a reason field and a
   step-up password, whose min-content is a real control box and not a character. The
   premise of the `nowrap` does not hold here, which is why the exemption is scoped to the
   same tables `width: auto` already names rather than widened.

   ⚠ MEASURED BOTH WAYS before it was believed: 1038 → 958 on the pending-cheque control
   and 1089 → 958 on the row's own screen, with the table then exactly filling its wrap. */
.vz-table--dues .vz-c-actions,
.vz-table--receipts .vz-c-actions,
.vz-table--collect .vz-c-actions,
.vz-table--runs .vz-c-actions,
.vz-table--onboarding .vz-c-actions { white-space: normal; }

@media (max-width: 64rem) {
  /* The schedule's own card layout. The generic phone rules above name the MEMBER
     columns, so without this a schedule row's cells would match no grid area and
     be auto-placed in an order nobody chose — the same trap `.vz-table--chits`
     exists for. */
  /* 🔴 `F47` — `dividend` SAT IN THE `1fr` TRACK BESIDE `net`, so the discount's right edge
     moved with whatever the net figure and the status chip needed. **7px of spread over
     twenty rows** — small, and the same defect `--collect` shows at 89px: an edge nobody
     chose. Both money lines span now, so both end where the card does. */
  .vz-table--schedule tr {
    grid-template-areas:
      'no       status'
      'when     when'
      'dividend dividend'
      'net      net'
      'actions  actions';
  }

  .vz-c-no       { grid-area: no; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-c-when     { grid-area: when; white-space: normal; }
  /* ⚠ `width: auto` UNDOES THE SHRINK-TO-FIT: this is a grid area now, not a table
     cell, and 1% of a card is 3px.

     🐛 ⚠ AND `white-space: normal` IS THE OTHER HALF, MEASURED RATHER THAN REASONED.
     The base rule is `nowrap`, which is right for a table cell and wrong for a card:
     the moment the members list carried THREE labelled actions, the row could not
     break and pushed the page **99px sideways at 320px**. The roster has been a
     wrapping flex row since B5 for exactly this reason; that treatment is the
     general one and this is where it belongs.

     ⚠ THE SIBLING MARGIN GOES WITH IT. `gap` and `margin-left` would both apply and
     the spacing would double on every action but the first. */
  .vz-c-actions {
    grid-area: actions;
    display: flex;
    flex-wrap: wrap;
    gap: var(--vz-space-2);
    width: auto;
    white-space: normal;
    text-align: left;
  }

  .vz-c-actions .vz-action--row + .vz-action--row { margin-left: 0; }

  /* 🐛 ⚠ `:first-of-type` MATCHED NEITHER OF THESE, AND BOTH FIGURES LANDED ON TOP OF
     EACH OTHER. This is the THIRD sighting of one mechanism: C2c corrected it on
     `.vz-table--runs` and wrote down why, `§D8`'s note repeated the rule — *"two money
     cells in one row need two classes"* — and the schedule was missed by both.

     `:first-of-type` matches by ELEMENT type: the first `<td>` among its siblings, which
     here is `.vz-c-no`. `:last-of-type` is `.vz-c-actions`. So neither selector ever
     matched, both money cells fell through to the unscoped `.vz-c-money { grid-area:
     value }` — an area THIS TEMPLATE DOES NOT DEFINE — and were auto-placed into an
     implicit row below `actions`, overlapping. A phone read **`₹₹500.00`** where the
     dividend and the net subscription should have been, on the schedule a collector
     opens, while the `dividend net` row of the template stayed empty.

     It measured ZERO overflow at every width from 320 to 1440, returned 200, and had a
     green wire test. It was found by reading the computed grid areas.

     `.vz-c-total` is the second money class the vocabulary already carries —
     `.vz-table--dues`, `--collect` and `--arrears` each pair it with `.vz-c-money` for
     exactly this reason. `testNoGridAreaIsAssignedThroughAnOfTypeSelector` now fails the
     suite on the mechanism rather than on this one instance. */
  /* ⚠ NOTHING SETS AN ALIGNMENT HERE, AND THAT IS DELIBERATE. `.vz-c-money`'s own card
     rule already carries `justify-self: end`, so each figure shrink-wraps and sits at the
     end of its own track — two money amounts ending on their column's right edge, which is
     what a column of figures does everywhere else in the product. A `text-align` on a box
     that is exactly its content wide would be a declaration with no effect. */
  .vz-table--schedule .vz-c-money { grid-area: dividend; }
  /* ⚠ `justify-self: end` IS GONE AND THE SHARED `F46`/`F47` RULE REPLACES IT (`justify-self:
     stretch`). The comment above is right that an alignment inside a shrink-wrapped box does
     nothing — which is exactly why the box is no longer shrink-wrapped. */
  .vz-table--schedule .vz-c-total { grid-area: net; }
}


/* ===========================================================================
   Enrolments: the roster and the module list (B5)
   ---------------------------------------------------------------------------
   Two tables, one vocabulary. Nothing here introduces a colour — every value is
   a `--vz-*` token, and none of it needs JavaScript.
   =========================================================================== */

/* The ticket number is the seat's identity — printed on receipts, read into the
   minutes, and the order rounding crumbs are distributed in. Tabular figures so a
   column of them lines up and 1 cannot be misread for 7. */
.vz-c-no,
.vz-c-share {
  font-variant-numeric: tabular-nums lining-nums;
  white-space: nowrap;
}

.vz-c-share { color: var(--vz-muted); }

/* ⚠ THE GROUP CODE IS A LINK SITTING ALONE IN A CELL, so the inline-text exemption
   does not cover it — WCAG 2.5.8 wants 24×24 and the project's own standard says to
   fix it rather than argue the exemption. The same rule `.vz-c-phone a` and
   `.vz-c-name a` already carry, for the same reason.

   Found by MEASURING at 375px, not by looking: the link renders at 68×17 and nothing
   about the page appears wrong. B4 fixed the identical defect on a page subtitle. */
.vz-c-code a {
  display: inline-flex;
  align-items: center;
  min-height: 24px;
  color: inherit;
}

.vz-c-code a:hover { text-decoration: underline; }

.vz-c-code a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* A roster row carries several actions rather than one, so they wrap instead of
   forcing the table wider. `.vz-c-actions` keeps `white-space: nowrap` for the
   single-action tables that came before it. */
/* 🐛 ⚠ THE FLEX MOVED OFF THE `<td>` AND ONTO AN INNER `<div>` (E6e), AND THAT IS THE
   TABLE DEFECT THE OWNER PHOTOGRAPHED. This rule used to be
   `.vz-table--roster .vz-c-actions { display: flex }` — **a table cell set to flex stops
   being a table-cell**: it no longer stretches to the row's height, so its
   `border-bottom` drew 19px above every other cell's and the row separator visibly broke
   under the ACTIONS column. Measured on a 22-seat roster: five cells ending at y=394 and
   that one at y=375.

   ⚠ AND IT CAUSED THE SECOND SYMPTOM TOO. A wrapping flex cell has a tiny min-content
   width, so the table algorithm collapsed it under any pressure — 135px wide, three
   actions stacked, a **165px row for one line of data**. `§E6c` exempted the roster from
   `width: 1%` to stop a related squash and left this in place; with the flex on a `<div>`
   the cell is an ordinary table-cell again and needs no exemption at all.

   🐛 ⚠ AND `nowrap` IS THE HALF THAT ACTUALLY FIXES THE ROW HEIGHT — moving the flex to a
   `<div>` alone did NOT. Measured after that change: the seam was gone and the row was
   still **157px**, because a wrapping flex row's min-content width is ONE glyph, so the
   table algorithm went on compressing the column to 87px and stacking three icons.
   `nowrap` makes the min-content the whole row of glyphs, so the table has to give it
   that width — and when the table then exceeds its box, **`.vz-table-wrap` scrolls,
   which is what it is for**. A cell that silently reflows instead of scrolling is the
   table quietly choosing a worse answer than the one the wrapper already provides.

   ⚠ THE CARD LAYOUT WRAPS, AND SETS THAT ITSELF. Below 44rem there is no table
   algorithm to compress anything and the actions carry their words again, so wrapping
   is right there and only there. */
/* 🔴 THE CONTROLS WRAP SINCE ITEM 63's ROW, WHICH IS `.vz-c-actions`' CHANGE ONE LEVEL DOWN.
   ⏸ `flex-wrap: nowrap` meant three 44px icon buttons were one 144px line that could never
   break, so `/app/users` could not get below **818px in a 728px wrap** however its data columns
   were treated — the action cell alone was 176 of it. ▶ Letting them stack closed it: measured
   at 1025, **47px → 0**.
   ⚠ **THE CELL ALREADY SAID `white-space: normal` AND IT CHANGED NOTHING**, because these are
   flex items rather than text — which is the trap worth naming: relaxing the text rule on a
   flex container looks like it should help and cannot. */
.vz-row-actions {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-1);
  justify-content: flex-end;
  white-space: normal;
}

/* 🔴🔴 AND WHERE THERE IS ROOM FOR ONE LINE, THE GLYPHS TAKE ONE LINE (owner-reported
   2026-09-16, `§7` item 101). The owner's screenshot showed the members register's three
   controls — view, edit, pause — stacked one per line in an 87px column at a DESKTOP width,
   with roughly 250px of slack sitting unused beside them.

   🔴 THE SWEEP COULD NOT SEE IT, AND THAT IS THE POINT. `documentElement.scrollWidth -
   clientWidth` read **0 at every width** the whole time, because nothing was overflowing:
   the column had simply collapsed. Measured at 1440 as shipped: the cell is **87 × 149**
   with three 44px controls at y = 404, 448, 492. `§E4`'s rule — a zero-overflow sweep is
   not the same as the screen being right — and the roster is worse at **87 × 253**.

   ▶ THE CAUSE IS `width: 1%` MEETING CONTENT THAT MAY WRAP. The shrink-to-fit idiom asks
   the table algorithm for the column's MIN-CONTENT, and the min-content of a wrappable row
   of glyphs is ONE GLYPH. So the column is one glyph wide at every viewport width, and the
   slack the idiom hands to the data columns is paid for in row height forever.

   ⚠ THE COMMENT BLOCK ABOVE PREDICTED THIS EXACTLY AND WAS LEFT ASSERTING THE OPPOSITE.
   It still reads *"`.vz-c-actions` is `white-space: nowrap`"* and explains that relaxing it
   means *"the table algorithm squashes the column and stacks the icons"* — written while
   item 63's row was in the act of relaxing it. That row measured the overflow it was
   fixing (members 83px → 0 at 1025) and did not measure the cell it was reshaping.
   ⛔ SO THIS IS NOT A REVERSAL OF THAT ROW EITHER: below the breakpoint the wrap still
   stands, which is what keeps the register from scrolling in the band item 63 cares about.

   🔴 BOTH PATTERNS NEED NAMING, AND ONE OF THEM IS NOT `white-space` AT ALL. Twenty-eight
   views put the controls straight into the cell, where `white-space` governs; six wrap them
   in `.vz-row-actions`, which is a FLEX box, where only `flex-wrap` does. Measured on the
   staff register: `nowrap` on the cell alone changed **nothing** (87 × 108 before and
   after) until `flex-wrap` was named too.

   ⏸ THE BOUND WAS 70rem AND IT IS NOW THE CARD BREAKPOINT (`§7` item 90, owner 2026-09-17:
   *"Each table record has own actions and actions icons should not go next line."*). The
   70rem version left a band — **1025–1119** — where the cell still stacked, bought by not
   letting the register scroll. **The owner was shown that trade and reversed it**: a
   horizontally scrolling table inside `.vz-table-wrap`, in a 78px band, beats a row five
   times taller than its data at every width in it.

   🔴 SO THE PRICE IS REAL AND IT IS RECORDED RATHER THAN DISCOVERED LATER. Item 101
   measured the minimum viewport each register needs before its table scrolls with one-line
   actions — **staff 1025, roster 1053, members 1103** — so from here the roster scrolls
   1025–1052 and the members register scrolls 1025–1102, in the wrap that exists for it,
   with page overflow still 0. ⚠ **That band is not exotic: 1024 CSS px is a 1280 laptop at
   125% Windows scaling.** ⛔ It is a deliberate PARTIAL reversal of item 63, which bought
   those widths by letting the cell wrap.

   ⚠ 64.0625rem IS ONE PIXEL ABOVE THE CARD BREAKPOINT, WHICH IS THE WHOLE POINT OF THE
   NUMBER. `@media (max-width: 64rem)` above turns every row into a card, and the two ranges
   have to be adjacent and non-overlapping: `min-width: 64rem` would match at exactly 1024
   as well, where the later rule would win and the card's own wrap would be lost.

   ⛔ AND THE RULE STOPS AT THE CARD RATHER THAN APPLYING "AT ANY WIDTH" LITERALLY, BECAUSE
   THE CARD IS NOT A TABLE ROW. The actions there are a full-width band under the data, not
   a column beside it, and the card block above records what forcing them onto one line
   costs: *"the moment the members list carried THREE labelled actions, the row could not
   break and pushed the page 99px sideways at 320px."* ▶ The owner's complaint is about a
   column collapsing next to unused slack; a card has neither.

   ✅ MEASURED BOTH WAYS ON THREE REGISTERS, with the table exactly filling its wrap and page
   overflow 0 at every step: members **149 → 61**, staff **108 → 60**, roster **253 → 79**.

   ⛔ THE FIVE FORM-HOLDING TABLES KEEP WRAPPING, AND BY SPECIFICITY RATHER THAN BY REPETITION.
   `.vz-table--dues .vz-c-actions` is (0,2,0) and beats the (0,1,0) rule below it whatever the
   media query says, so the `white-space: normal` they were each given above still wins. ⚠ None
   of the five uses `.vz-row-actions` today — their cells hold whole forms, not glyph rows — so
   `flex-wrap` needs no exemption yet. A sixth table that holds a form AND uses the wrapper must
   join both lists, which is the same membership rule stated above. */
@media (min-width: 64.0625rem) {
  .vz-c-actions { white-space: nowrap; }
  .vz-row-actions { flex-wrap: nowrap; }
}

/* ⚠ WIDENED FROM `.vz-table--roster` TO EVERY ACTION CELL (E6b). A `<form>` is a block,
   so a glyph inside one drops to its own line — **measured 105px tall on the members
   list**, where Pause is a POST beside two links. The roster found this first and the
   rule was scoped to the one table that had hit it; the members list is the second, and
   the reason was never roster-specific.

   The original note, kept: the remove button is a whole form; without this it becomes a
   block and pushes the
   other actions onto their own line at every width. */
.vz-c-actions form { display: inline-flex; }

@media (max-width: 64rem) {
  /* ⚠ EACH TABLE NEEDS ITS OWN CARD LAYOUT. The phone rules above turn any
     `.vz-table` row into a card using `grid-template-areas` that name the MEMBER
     columns; a roster row's cells match none of them, get auto-placed into implicit
     rows, and come out in an order nobody chose. The same trap `.vz-table--chits`
     and `.vz-table--schedule` exist for.

     Nothing is hidden at 375px, which is the point: staff on a phone are the ones
     asking "is that group full yet", and a column that vanishes is the answer they
     came for. */
  .vz-table--roster tr {
    grid-template-areas:
      'name    status'
      'no      share'
      'joined  joined'
      'actions actions';
  }

  /* ⚠ `actions` IS A NAMED AREA HERE FOR THE SAME REASON EVERY OTHER CELL IS (E6b).
     An unnamed cell is auto-placed in an order nobody chose and still measures zero
     overflow, because it fits. */
  .vz-table--enrollments tr {
    grid-template-areas:
      'name    status'
      'code    no'
      'share   joined'
      'actions actions';
  }

  /* ⚠ THE PROFILE'S SEATS PANEL (E6d). Its own areas for the reason every table here
     has them: an unnamed cell is auto-placed in an order nobody chose and still
     measures zero overflow, because it fits. */
  /* 🐛🔴 A ONE-COLUMN CARD SINCE ITEM 64's FIX, AND THE MEASUREMENT IS WHY.
     ⏸ It paired `name` with `status`, and `status` holds an arrears chip whose `auto` track
     takes its max-content — so at 320 the tracks computed to **`30.11px 154.89px`** and the
     group name had **30px** to render "Settlement Wire Group" in. ▶ That is `§7` item 64's own
     sentence — *"the second column is sized by its content and the first may shrink"* — and its
     own prescription: **a row whose second cell is a chip should not be sharing a line with
     the thing that names it.** `.vz-table--drift` is the precedent, from `§Q3`.
     ⚠ **THIS IS A SECOND DEFECT ON THE SAME TABLE AND THEY ARE NOT THE SAME ONE.** The first
     was a modifier COLLISION that left two cells unplaced (see `--followup-seats` below);
     fixing it revealed this, which was underneath all along and invisible while the wrong
     block was winning. **Measured after the rename, not before.** */
  .vz-table--seats tr {
    grid-template-areas:
      'name'
      'no'
      'share'
      'status'
      'standing'
      'actions';
    grid-template-columns: minmax(0, 1fr);
  }

  /* 🐛 ⚠ AND THE CELL HAS TO BE ASSIGNED TO THE AREA, NOT JUST NAMED IN THE TEMPLATE
     (E6d). `standing` was in `grid-template-areas` above and `.vz-c-standing` had no
     `grid-area` at all, so it was auto-placed into an implicit row and widened the
     card — **32px of page overflow at 320px**. Caught by the sweep reading the
     computed `grid-area` of every cell, which is `§E4` constraint 7's *"read the
     computed grid areas, not just the overflow number"* finding its own defect the
     first time it was pointed at a brand-new table. */
  .vz-c-standing { grid-area: standing; }

  .vz-table--seats .vz-c-no { justify-self: start; }

  .vz-c-no     { grid-area: no; }
  .vz-c-share  { grid-area: share; justify-self: end; }

  .vz-table--enrollments .vz-c-no { justify-self: end; }
  .vz-table--enrollments .vz-c-share { justify-self: start; }

  .vz-table--roster .vz-c-actions,
  .vz-row-actions {
    grid-area: actions;
    justify-content: flex-start;
    /* Words are back at this width, and a card has no table algorithm to fight. */
    flex-wrap: wrap;
  }

  /* Full-width targets on the surface where they are actually tapped. */
  .vz-table--roster .vz-c-actions .vz-action {
    min-height: var(--vz-tap-min);
  }
}


/* ===========================================================================
   Auctions: the branch diary (D4 / C1b)
   ---------------------------------------------------------------------------
   ⚠ THE MODIFIER IS LOAD-BEARING, NOT DECORATION. At phone width the generic
   rules turn a table row into a CSS grid whose areas are named per modifier —
   for the MEMBER columns. A table shipped without its own modifier has cells
   matching no area at all, so they are auto-placed in an order nobody chose,
   and the horizontal-overflow measurement still reads ZERO because they fit.
   That is exactly how B5 shipped two broken tables, so this block goes in with
   the markup rather than after somebody notices.
   =========================================================================== */

/* 🔴 NEITHER OF THESE IS A DICTATED VALUE, SO NEITHER KEEPS `nowrap` (item 63's row).
   ⏸ Both carried `white-space: nowrap` with no reason recorded beside it. The reason this file
   gives for `nowrap` everywhere it IS justified is always the same one — `.vz-c-code` and
   `.vz-c-phone` are *"read digit by digit and dictated over the phone"*, and `.vz-audit-values`
   says *"a value split across two lines is a value somebody copies wrongly"*. **A winner is a
   person's name and a status is a badge**; nobody copies either off the screen.

   ⚠ AND THE BADGE ITSELF DOES NOT BREAK. `.vz-badge` is `inline-flex`, so what wraps is the
   badge AWAY from any note beside it, never the word inside it.

   Measured at 1025: `--settlements` **74px** narrower from the status column alone, `--roster`
   closed outright, `--dunning` 9px. */
.vz-c-winner,
.vz-c-status { white-space: normal; }

/* 🐛 VAZHAI'S SUPPORT QUEUE AS A CARD (found driving, 2026-11-18). It shipped without its own
   template, and BOTH the operator and the subject are `.vz-c-name`, so on a phone the two were
   drawn in the generic card's one `name` area, ON TOP OF EACH OTHER — "Kovai Chit Funds" over
   "Coimbatore receipt printer jammed", unreadable, at zero page overflow. It predates the branch
   column; "operator · branch" made it longer, and the drive caught it. The card leads with what
   was asked and its state, then who asked and from which branch, then the two clocks. */
@media (max-width: 64rem) {
  .vz-table--support-queue tr {
    grid-template-areas:
      'name      status'
      'operator  operator'
      'code      priority'
      'response  resolution';
  }

  .vz-table--support-queue .vz-c-operator   { grid-area: operator; font-size: var(--vz-fs-sm); color: var(--vz-muted); }
  .vz-table--support-queue .vz-c-priority   { grid-area: priority; justify-self: end; }
  .vz-table--support-queue .vz-c-response   { grid-area: response; }
  .vz-table--support-queue .vz-c-resolution { grid-area: resolution; justify-self: end; }
}

@media (max-width: 64rem) {
  /* 🔴 THE REGISTER'S CARD, AND IT WAS THE DIARY'S UNTIL `§7` ITEM 109 SPLIT THEM.
     The diary dropped Status, Winner and Discount — they were constant by construction
     once a held month left the screen — and took `.vz-table--diary` with it. ⛔ THIS BLOCK
     IS DELIBERATELY UNCHANGED: on the register a winner and a discount are the point, and
     leaving the rule untouched is what makes *"the register did not move"* provable by a
     before/after capture rather than argued in a comment.

     ⚠ WHY A SECOND TEMPLATE RATHER THAN THREE UNUSED AREAS. `grid-template-areas` declares
     five area ROWS here. A table that stops filling two of them still pays their two grid
     GAPS — dead space on a phone card, **at zero page overflow**, which is this file's own
     named blind spot arriving through the layout instead of through a missing class.

     The card leads with when and which group — the two things staff scan for — and puts
     the outcome underneath. */
  .vz-table--auctions tr {
    grid-template-areas:
      'when     no'
      'group    group'
      'status   discount'
      'winner   winner'
      'actions  actions';
  }

  .vz-table--auctions .vz-c-when   { grid-area: when; white-space: normal; }
  .vz-table--auctions .vz-c-no     { grid-area: no; justify-self: end; }
  .vz-table--auctions .vz-c-name   { grid-area: group; }
  .vz-table--auctions .vz-c-status { grid-area: status; }
  .vz-table--auctions .vz-c-money  { grid-area: discount; justify-self: end; }
  .vz-table--auctions .vz-c-winner { grid-area: winner; }

  .vz-table--auctions .vz-c-actions {
    grid-area: actions;
    text-align: left;
  }

  /* A full-width tap target on the surface where it is actually tapped —
     `--vz-tap-min` is the 24px floor the responsive pass measures against. */
  .vz-table--auctions .vz-c-actions .vz-action {
    min-height: var(--vz-tap-min);
  }
}

/* ===========================================================================
   The auctions DIARY's card (`§7` item 109)
   ---------------------------------------------------------------------------
   Four columns — Due, Group, Inst., Actions — because the other three were
   constant by construction: once a held month leaves this screen, every row on
   it is `pending` or `scheduled`, so Status always read "Not held" and Winner
   and Discount were always an em dash. Owner-answered 2026-09-18, option ⓶.

   ⚠ THE REGISTER IS NOT THIS TABLE. It wears `.vz-table--auctions` above, keeps
   all seven columns, and is untouched by this block — the two screens ask
   different questions and one card template can no longer answer both.
   =========================================================================== */

@media (max-width: 64rem) {
  /* Three area rows, not five. The two the diary no longer fills would each
     still take their grid gap if this reused the register's template. */
  .vz-table--diary tr {
    grid-template-areas:
      'when     no'
      'group    group'
      'actions  actions';
  }

  .vz-table--diary .vz-c-when { grid-area: when; white-space: normal; }
  .vz-table--diary .vz-c-no   { grid-area: no; justify-self: end; }
  .vz-table--diary .vz-c-name { grid-area: group; }

  .vz-table--diary .vz-c-actions {
    grid-area: actions;
    text-align: left;
  }

  /* A full-width tap target on the surface where it is actually tapped —
     `--vz-tap-min` is the 24px floor the responsive pass measures against. */
  .vz-table--diary .vz-c-actions .vz-action {
    min-height: var(--vz-tap-min);
  }
}

/* 🔴 HOW FAR BEHIND AN OVERDUE DRAW IS, UNDER THE DATE IT WAS DUE (`§7` item 103).
   The diary's overdue bucket is the only alarm on the screen, so the sub-line under
   the date carries urgent ink rather than `.vz-sub-meta`'s subtle grey.

   ⚠ IT IS SPELLED OUT HERE RATHER THAN LEFT TO `.vz-tone--urgent` ALONE, AND THAT IS
   ITEM 101's FINDING APPLIED BEFORE IT COULD BITE. `.vz-sub-meta { color: var(--vz-subtle) }`
   and `.vz-tone--urgent { color: var(--vz-danger-text) }` are BOTH (0,1,0), so which one
   paints the span depends only on which appears later in this file.
   ⚠ MEASURED, NOT ASSUMED: today `.vz-tone--urgent` DOES win — the two classes together
   outside this table read `#791F1F` in the rendered capture — so this rule changes nothing
   right now. It is here because the thing that makes it true is source order, and source
   order is what item 101's fix lost at equal specificity while looking fixed on the one
   screen anybody checked. (0,3,0) cannot be resolved that way.

   ⚠ IT IS `--diary` SINCE `§7` ITEM 109, NOT `--auctions`. The days-late line is the
   diary's alarm and the REGISTER HAS NEVER RENDERED ONE — grepped, not assumed — so a
   rule left on the shared modifier would have been a rule for nobody. */
.vz-table--diary .vz-c-when .vz-tone--urgent {
  color: var(--vz-danger-text);
}


/* ===========================================================================
   Collections: what a seat owes, and what it has paid (C2b)
   ---------------------------------------------------------------------------
   Two tables and one inline form. Nothing here introduces a colour — every
   value is a `--vz-*` token — and none of it needs JavaScript.
   =========================================================================== */

/* Money columns are tabular so a column of figures lines up and a member can
   run a finger down it. The same rule `.vz-c-no` already carries, for the same
   reason: these are read aloud and checked by hand at a counter. */
/* ⚠ F38 TOOK `text-align: right` OUT OF HERE TOO — see the base table section. */
.vz-c-penalty,
.vz-c-total,
.vz-c-money {
  font-variant-numeric: tabular-nums lining-nums;
  white-space: nowrap;
}

/* An accrued late fee that is zero is the ordinary case, and rendering it as
   loud as the principal beside it makes every row look like a problem. */
.vz-c-penalty { color: var(--vz-muted); }

/* ⚠ A TYPE WORD, NOT A VALUE — so it wraps (item 63's row, and `.vz-c-status`' reasoning one
   section up). This column holds a settlement type ("Member exit", "Foreman final") and a
   collector's name; both are prose. Measured at 1025: 14px off `--settlements`. */
.vz-c-mode { white-space: normal; }

/* The retraction controls sit in a table cell, so they wrap rather than push
   the row wide. `column-gap` keeps the reason, the password and the button from
   touching at any width. */
.vz-row-form {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-2);
}

/* ⚠ A SHORT CONTROL IS STILL A 24px TAP TARGET. B5 found the identical defect on
   a link sitting alone in a cell — WCAG 2.5.8 wants 24×24 and this project's
   standard is to fix it rather than argue the exemption. */
.vz-control--inline {
  width: auto;
  min-width: 8rem;
  min-height: var(--vz-tap-min);
}

@media (max-width: 64rem) {
  /* ⚠ WITHOUT THESE, EVERY CELL BELOW MATCHES NO GRID AREA and is auto-placed
     in an order nobody chose — and it measures ZERO overflow because it still
     fits, which is exactly the blind spot B5 found and C1b re-checked by
     reading the computed areas rather than the overflow number. The generic
     phone rules above name the MEMBER columns, so a dues row would inherit a
     layout meant for a different table. */
  /* 🔴 `F47` — `principal` SAT IN THE `1fr` TRACK, so its right edge was the card's right
     edge minus whatever `when` or `penalty` happened to be wide. The fixture renders this
     table one row at a time, so no spread could be MEASURED here — **the grid is the
     evidence, not the capture**, and it is the same shape `--collect` was measured on. */
  .vz-table--dues tr {
    grid-template-areas:
      'no        when'
      'principal principal'
      'penalty   penalty'
      'total     total';
  }

  .vz-table--dues .vz-c-no    { grid-area: no; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--dues .vz-c-when  { grid-area: when; white-space: normal; justify-self: end; text-align: right; }
  .vz-table--dues .vz-c-money { grid-area: principal; }
  .vz-table--dues .vz-c-penalty { grid-area: penalty; }
  .vz-table--dues .vz-c-total { grid-area: total; }

  /* The footer's total is a two-cell row of its own and must not inherit the
     body's areas, or the label and the figure land on top of each other. */
  .vz-table--dues tfoot tr {
    grid-template-columns: minmax(0, 1fr) auto;
    grid-template-areas: 'label total';
  }

  .vz-table--dues tfoot th { grid-area: label; }

  .vz-table--receipts tr {
    grid-template-areas:
      'code    status'
      'when    mode'
      'amount  amount'
      'actions actions';
  }

  .vz-table--receipts .vz-c-code   { grid-area: code; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--receipts .vz-c-status { grid-area: status; justify-self: end; }
  .vz-table--receipts .vz-c-when   { grid-area: when; white-space: normal; }
  .vz-table--receipts .vz-c-mode   { grid-area: mode; justify-self: end; }
  .vz-table--receipts .vz-c-money  { grid-area: amount; text-align: left; }

  .vz-table--receipts .vz-c-actions {
    grid-area: actions;
    text-align: left;
  }

  .vz-table--receipts .vz-c-actions .vz-action,
  .vz-table--receipts .vz-c-actions .vz-control--inline {
    min-height: var(--vz-tap-min);
  }

  /* On a phone the reason and the password each take the full width; side by
     side they would each be too narrow to type a sentence into. */
  /* On a phone the reason and the password each take the full width; side by
     side they would each be too narrow to type a sentence into. */
  .vz-row-form .vz-control--inline { flex: 1 1 100%; }

  /* 🔴 F41 — AND THE EXCEPTION BELOW IT, FOR A PANEL ROW THIS WAS NEVER MEANT TO REACH.
     The rule above is written for a row-form in a TABLE CELL, and 64rem is where those
     become phone cards. ⛔ `members/show.php`'s surety row is not a table — it is a
     `.vz-sub-row` in a panel — so at **1024px**, a laptop, the reason field went full-width
     and dropped *Release* onto a second line **with 246px of empty space to the row's
     right**. That is the raggedness `F41` was raised for.
     ▶ **ADDITIVE AND SCOPED TO `.vz-sub-actions` (0,3,0), NOT A RE-SCOPING OF THE RULE
     ABOVE.** The three row-forms in a table cell — `collect`, `collect-batch`, `messaging` —
     are untouched **by construction**, which matters because the drive fixture renders none
     of those three states and none of them could be verified by looking.
     ⚠ **THE FIELD STILL GETS ITS OWN LINE WHEN IT NEEDS ONE**: measured, input and button sit
     together down to 375px and the button wraps beneath it at 320 on its own, from
     `.vz-row-form`'s own `flex-wrap: wrap`. **Nothing here forces a single line.**
     ⚠ **AND THE TWO ARE ASSERTED ADJACENT** by `testThePanelActionRowCanShrinkAndKeepsItsField
     BesideItsButton`, so the exception cannot drift away from the rule it excepts. */
  .vz-sub-actions .vz-row-form .vz-control--inline { flex: 0 1 auto; }
}


/* ===========================================================================
   Bringing in a book — the import screen's own layout (Phase M, M3; `§Q-R5`)
   ---------------------------------------------------------------------------
   🔴 THIS WHOLE SECTION USED TO LIVE INSIDE `@media (max-width: 44rem)`, WHICH IS
   WHY THE SCREEN HAD NO DESKTOP LAYOUT AT ALL. M3 wrote the three card layouts,
   measured them, and stopped — so above 44rem every cell fell back to the generic
   `.vz-table` rules and the three defects the owner reported on 2026-09-15
   (*"Bring in a book UI is not build good. Lot of alignment issues."*) were all
   one missing half of one file.

   ⚠⚠ AND NO TEST COULD SEE ANY OF IT. `.vz-table-wrap` clips, so
   `documentElement.scrollWidth - clientWidth` read **0px at all 1,121 widths from
   320 to 1440** on every one of the five captured states, before and after. Every
   number in the comments below is `table.scrollWidth` against `wrap.clientWidth`,
   or a box read off the cell itself. This is `§7` item 63's family.
   =========================================================================== */

/* ── 1. THE LISTS TABLE, IN TABLE MODE ──────────────────────────────────────

   🐛 THE COLUMN CONTRACT IS THE LONGEST CONTENT ON THE PAGE AND IT IS A `<code>`
   BLOB, so the cell that holds it is the one that decides this table's width.
   `white-space: normal` was phone-only and the base cell is already `normal`, so
   that half was never the desktop problem; `overflow-wrap: anywhere` was, because
   `installments_elapsed_at_cutover` is 31 characters with no break opportunity in
   it and a comma-separated list gives the algorithm nothing else to work with.

   ⚠ IT IS SAFE HERE ONLY BECAUSE THE THREE COLUMNS BESIDE IT HAVE FLOORS.
   `§E6`'s standing warning is that `anywhere` takes a column's min-content down to
   ONE CHARACTER, so a cell carrying it can be crushed to a vertical stack of
   letters at some width in the middle of the range, at zero page overflow. Measured
   after this change: the column holds **227px at 1049** and **409px at 1280**, and
   its narrowest anywhere above the card breakpoint is **206px at 1025** — because
   `list`, `rows` and the upload form each take a real minimum out of the table
   first, so there is never enough slack for this one to be crushed. */
.vz-table--onboarding .vz-c-columns {
  white-space: normal;
  overflow-wrap: anywhere;
}

/* 🐛 THE BLANK-FILE LINK WAS FLOATING MID-SENTENCE, AND THAT IS THE OWNER'S SECOND
   DEFECT. The `<code>` blob is an inline box and the link beside it is
   `display: inline-flex` with a 44px tap floor, so the two shared the last line of
   the column contract: measured at 1049, the link sat at **x 544 inside a cell
   running 469→857 with its top at 453 while the code ran to 483** — a button
   jammed into the middle of a paragraph of column names.

   ▶ THE FIX IS THE STRUCTURE, NOT A MARGIN. The code becomes a block, so the link
   has a line of its own underneath it; re-measured, `link.top >= code.bottom` at
   every width. ⛔ A FIFTH COLUMN WAS THE OTHER OPTION AND IS WRONG: this table
   already scrolled its wrap across 173 widths, and a rigid column would have been
   paid for out of the contract cell that is the reason the screen exists. */
.vz-table--onboarding .vz-c-columns code { display: block; }

.vz-table--onboarding .vz-c-columns .vz-action { margin-top: var(--vz-space-2); }

/* 🐛🔴 THE UPLOAD CONTROL'S FLOOR, AND IT IS THE HALF `width: auto` DOES NOT BUY.
   Joining the `width: auto` list above took the actions column from 135px to 157
   and the file input from 26px to 49 — better and still clipped, because the input
   is a flex item in an `inline-flex` form and a native file input's own min-content
   is far below the **219px** it needs to draw its own *"Choose file · No file
   chosen"*.

   ⚠ SO THE FLOOR IS ON THE CONTROL, AND `§Q-H7`'s RULE IS THAT IT REALLY DOES FLOOR
   THE COLUMN: an `<input>` contributes an intrinsic min-content width to the table
   algorithm whatever `min-width: 0` says about flex shrinking. That is affordable
   **only because the card breakpoint below moved to 64rem** — the band where four
   content columns plus a form did not fit is now cards, and above 1024 the wrap is
   687px or wider.

   🐛 AND THE FORM WRAPS, WHICH IS WHAT MAKES THE FLOOR CHEAPER THAN THE CLIPPING
   WAS. `.vz-c-actions form` is `inline-flex`, so input and button on one line put
   **both** their widths into the column minimum: measured 287px at 1049, taken
   straight out of the column contract beside it, with the input STILL truncating
   its status text to *"N…en"*. Letting the form wrap and giving the input its own
   line makes the minimum the input alone. Measured at 1049: actions **258px**
   (down from 287), input **226px** with its whole label drawn, the contract column
   **227px** (up from 229 at a narrower `rows` column), and the table exactly
   filling its 710px wrap. ⛔ The one-line layout and a legible control were not
   both available at this width.

   ⚠⚠ AND THE FULL NATIVE LABEL IS NOT AVAILABLE HERE AT ALL — MEASURED, SO THAT
   THE NEXT ROW DOES NOT GO LOOKING FOR IT. A `.vz-control` file input needs
   **320px (20rem)** before Chrome stops ellipsising *"No file chosen"* to
   *"No fil…osen"*; at 1049 that would floor the actions column at 352 out of a
   710px wrap and leave the column contract — the reason this screen exists — about
   130px. 13.75rem draws the *Choose file* button in full and ellipsises only the
   status text, which this row already states beside it in the *Rows* column as
   *"Not sent yet"*. ⛔ Styling the control out of its native shell is a component,
   not a rule, and `§M1` Q14's answer is why the upload is a plain file input. */
.vz-table--onboarding .vz-c-actions form { flex-wrap: wrap; gap: var(--vz-space-2); }

.vz-table--onboarding .vz-c-actions .vz-control {
  min-width: 13.75rem;
  flex: 1 1 100%;
}

/* 🐛 *"Not sent yet"* WAS RENDERING AS THREE STACKED WORDS BESIDE A ONE-LINE NAME,
   which is the owner's *"alignment issues"* in the plainest form the screenshot
   holds. The column is headed *Rows* and normally carries a figure; the empty case
   is a short phrase, and at 70px it broke after every word.

   ⚠ A `min-width` FLOOR RATHER THAN `white-space: nowrap`, DELIBERATELY. Both cost
   the same 32px today — the floor is what the phrase needs either way — but `nowrap`
   makes the string UNBREAKABLE, so a longer one in another script would widen the
   table instead of wrapping. `§E6` step 7 exists because Malayalam's compound words
   stress narrow columns harder than English does, and a rule that fails by wrapping
   is the one to leave behind.

   ⚠ 4.5rem IS THE CONTENT BOX AND THE CELL ADDS 32px OF PADDING ON TOP — measured,
   after a first draft at 6.5rem came out **136px wide** and quietly took 34px off
   the column contract beside it. `min-width` on a `td` floors the content, not the
   cell. */
.vz-table--onboarding .vz-c-rows { min-width: 4.5rem; }

/* The register of books, in TABLE mode (Phase M, M3).
   🔴 THESE TWO LINES ARE A MEASURED FIX, NOT A PRECAUTION. Six columns of real
   data pushed this table 30px past its wrap at 768 and 22px at 1024 — inside the
   wrap, so PAGE overflow read zero the whole time, which is `§6`'s *"measure the
   inner box, not only the page"* catching its fifth row running.

   ⚠ AND THE CONTROL IS WHAT SAID IT WAS A DEFECT RATHER THAN THE HOUSE NORM.
   `/app/collections` — a screen that row did not touch — measures **0 at 1024**
   with its own two registers, so a scrolling table here was mine to fix.

   A table cell is `nowrap`, which is right for a figure and wrong for a full
   timestamp and a sentence-shaped status. Letting exactly these two wrap takes
   the table back inside its wrap at every width without dropping a column or
   abbreviating an instant — and `§7` item 10's trap is why the instant is NOT
   shortened by hand: `created_at` is UTC and `DateHelper::display()` is what
   converts it to the operator's day.

   ⚠ KEPT WHEN THE BREAKPOINT MOVED, NOT SUPERSEDED BY IT. `§Q-R4`'s finding is
   that a structural fix and a content fix are not alternatives — the breakpoint
   closes 769→904 and this is what keeps the table inside its wrap at 1024, where
   it is a table again and the wrap is at its narrowest. */
.vz-table--onboarding-batches .vz-c-when,
.vz-table--onboarding-batches .vz-c-status {
  white-space: normal;
}


/* ── 2. ALL THREE TABLES BECOME CARDS AT 64rem, NOT 44 ──────────────────────

   🔴 THE BAND WAS MEASURED BEFORE THE BREAKPOINT WAS CHOSEN, ON THE REAL CAPTURES.
   With the default 44rem card switch, `table.scrollWidth` against
   `wrap.clientWidth` on the staged-book screen:

       .vz-table--onboarding          173 widths, 769 → 941   (601 in 430 at 769)
       .vz-table--onboarding-batches  136 widths, 769 → 904   (567 in 430 at 769)
       .vz-table--onboarding-rows     none

   ⚠ THE BAND STARTS AT 769 BECAUSE THAT IS WHERE THE 240px NAVIGATION ARRIVES
   (`.vz-nav` at 48rem) AND A TABLE DOES NOT BECOME A CARD UNTIL 44rem. The squeeze
   is structural rather than these tables' fault, which is why
   `.vz-table--billing-operators` and `.vz-table--subscription` hit it first and set
   this exact precedent. 64rem is reused rather than a third breakpoint invented:
   941 needs 60rem and 64 is already in this file.

   ⚠ AND THE CONTENT FIXES ABOVE DO NOT CLOSE IT ALONE — MEASURED, NOT ASSUMED.
   With every rule in section 1 applied and the breakpoint left at 44rem, the lists
   table still scrolled **143 widths, 769 → 911**. Both halves are here because
   each was swept separately.

   🔴 `--onboarding-rows` MOVES TOO, AND IT IS THE ONE WITH NO MEASUREMENT BEHIND
   IT. It fits at every width. It moves because all three tables are on ONE screen:
   at 900px the check results would have been the only panel still drawing a table
   while the lists above and the register below were cards, and a screen rendering
   two responsive idioms at one width is the owner's *"alignment issues"* in its
   own right. ⛔ That is a judgement, and it is recorded as one rather than dressed
   up as a measurement.

   ⚠ THE DISPLAY SWITCH IS RESTATED because the base one lives in the 44rem block
   and these tables need it 20rem earlier. Only these three modifiers are named, so
   no other table's breakpoint moves. */
@media (max-width: 64rem) {
  .vz-table--onboarding,
  .vz-table--onboarding tbody,
  .vz-table--onboarding tr,
  .vz-table--onboarding td,
  .vz-table--onboarding-rows,
  .vz-table--onboarding-rows tbody,
  .vz-table--onboarding-rows tr,
  .vz-table--onboarding-rows td,
  .vz-table--onboarding-batches,
  .vz-table--onboarding-batches tbody,
  .vz-table--onboarding-batches tr,
  .vz-table--onboarding-batches td {
    display: block;
    width: auto;
  }

  /* The header row still exists for assistive technology; it simply has nothing
     to label once the grid is gone. */
  .vz-table--onboarding thead,
  .vz-table--onboarding-rows thead,
  .vz-table--onboarding-batches thead {
    position: absolute;
    width: 1px;
    height: 1px;
    overflow: hidden;
    clip: rect(0 0 0 0);
    clip-path: inset(50%);
  }

  /* ⚠ THE CARD CHROME AND THE COLUMN COUNT ARE BOTH RESTATED. The generic card in
     the 44rem block draws the border, the padding and `minmax(0, 1fr) auto`; none
     of it reaches these tables between 44 and 64rem, and naming AREAS against a
     track list that is not there leaves auto-placement to invent one.
     `.vz-table--breakglass-roles` carries the longer version of this warning. */
  .vz-table--onboarding tr,
  .vz-table--onboarding-rows tr,
  .vz-table--onboarding-batches tr {
    display: grid;
    grid-template-columns: minmax(0, 1fr) auto;
    gap: var(--vz-space-1) var(--vz-space-3);
    margin-bottom: var(--vz-space-3);
    padding: var(--vz-space-4);
    border: 1px solid var(--vz-border);
    border-radius: var(--vz-radius-lg);
    background: var(--vz-surface);
  }

  .vz-table--onboarding tbody tr:hover td,
  .vz-table--onboarding-rows tbody tr:hover td,
  .vz-table--onboarding-batches tbody tr:hover td { background: none; }

  /* 🐛 `th` IS NAMED BESIDE `td` HERE AND IT IS NOT IN THE GENERIC CARD BLOCK, WHICH
     IS WHY THE LISTS CARD DREW A RULE UNDER EVERY LIST NAME. The name cell is a
     `<th scope="row">` — correct markup, and `.vz-table th` gives it the same
     `border-bottom` as a body cell, which the generic 44rem block then resets for
     `td` only. Inside a card that border is a horizontal line across the middle of
     the card that nobody chose. ⚠ IT IS PRODUCT-WIDE AND IS FIXED HERE ONLY FOR
     THESE THREE TABLES: nine views render a row header, and widening the generic
     rule would change eight screens this row has not captured. */
  .vz-table--onboarding th,
  .vz-table--onboarding td,
  .vz-table--onboarding-rows th,
  .vz-table--onboarding-rows td,
  .vz-table--onboarding-batches th,
  .vz-table--onboarding-batches td {
    padding: 0;
    border-bottom: 0;
  }

  /* 🔴 THESE EXIST BECAUSE THE SWEEP FOUND THEM AUTO-PLACED (M3). Below the card
     breakpoint `tr` becomes a grid and every cell was resolving to
     `grid-area: auto` — the card layout was arbitrary rather than designed, which
     is `§E4` constraint 7. Page overflow measured ZERO throughout, which is exactly
     why the areas have to be READ rather than the number trusted. */

  /* The lists table. ⚠ `columns` GETS ITS OWN FULL-WIDTH ROW: it holds the whole
     column contract as monospace text — the longest content on the page, and the
     only place a CSV operator can see what their file must contain, because
     `§M1` Q14 traded away the workbook's locked headers. Sharing a row with
     anything would make it the column that overflows the wrap. */
  .vz-table--onboarding tr {
    grid-template-areas:
      'list    rows'
      'columns columns'
      'actions actions';
  }

  .vz-table--onboarding .vz-c-list    { grid-area: list; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--onboarding .vz-c-rows    { grid-area: rows; justify-self: end; }
  .vz-table--onboarding .vz-c-columns { grid-area: columns; }
  .vz-table--onboarding .vz-c-actions { grid-area: actions; text-align: left; }

  /* 🔴 BOTH DESKTOP FLOORS COME OFF INSIDE A CARD, AND THIS IS NOT TIDYING.
     Measured: with the 13.75rem input floor left on, the card's own wrap reported
     the table scrolling at **320 through 324** — a 220px control inside a 197px
     card. A floor that exists to stop a table COLUMN collapsing has nothing to do
     on a card, where the cell already has the whole width; and the `rows` floor
     would take 104px of a 197px card out of the list name beside it. */
  .vz-table--onboarding .vz-c-actions .vz-control { min-width: 0; }
  .vz-table--onboarding .vz-c-rows { min-width: 0; }

  .vz-table--onboarding .vz-c-actions .vz-action,
  .vz-table--onboarding .vz-c-actions .vz-control {
    min-height: var(--vz-tap-min);
  }

  /* What the check found. ⚠ THE PROBLEM SENTENCE TAKES THE FULL WIDTH — it is a
     whole instruction ("Write the date as year-month-day, like 2026-03-05"), and
     the three identifiers beside it are short. */
  .vz-table--onboarding-rows tr {
    grid-template-areas:
      'list    line'
      'field   field'
      'problem problem';
  }

  .vz-table--onboarding-rows .vz-c-list    { grid-area: list; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--onboarding-rows .vz-c-line    { grid-area: line; justify-self: end; }
  .vz-table--onboarding-rows .vz-c-field   { grid-area: field; }
  .vz-table--onboarding-rows .vz-c-problem { grid-area: problem; white-space: normal; }

  /* The register of books. */
  .vz-table--onboarding-batches tr {
    grid-template-areas:
      'when    status'
      'cutover scope'
      'lists   rows';
  }

  .vz-table--onboarding-batches .vz-c-when    { grid-area: when; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--onboarding-batches .vz-c-status  { grid-area: status; justify-self: end; }
  .vz-table--onboarding-batches .vz-c-cutover { grid-area: cutover; white-space: normal; }
  .vz-table--onboarding-batches .vz-c-scope   { grid-area: scope; justify-self: end; }
  .vz-table--onboarding-batches .vz-c-lists   { grid-area: lists; }
  .vz-table--onboarding-batches .vz-c-rows    { grid-area: rows; justify-self: end; }
}


/* ===========================================================================
   The bulk collect and the collections worklist (C2c)
   ---------------------------------------------------------------------------
   Three more tables in the same vocabulary. No new colours, no JavaScript.
   =========================================================================== */

/* The tick column is the control, so it gets a real target rather than the 13px
   box a browser draws by default — WCAG 2.5.8's floor, and this is a control
   somebody uses twenty times in a row with a queue waiting. */
.vz-c-pick {
  width: 3rem;
}

/* ═══ 🔴 F39 — THE TICK IS CENTRED WHERE IT CAN WIN, AND ITS HEADER MOVES WITH IT ═════════
   Filed by F38's measurement, answered by the owner 2026-09-25: **turn this one on.** The
   `text-align: center` above was unprefixed at `0,1,0` and lost to `.vz-table th, .vz-table td`
   at `0,1,1`, so the tick has always rendered LEFT.

   ▶ **UNLIKE `.vz-c-actions`, THIS ONE HAS SOMETHING TO ALIGN.** Measured on
   `/app/chits/{id}/collections/{no}` at 1280px: the column is **108.7px** — its width comes
   from the word *"Collecting"*, not from the control — and the tick is **44px**, so there is
   **33px of slack** and centring moves it about 16px off the left edge.
   🔴 **AND THE `<th>` CARRIES `.vz-c-pick` NOW**, which is F38's pairing rule rather than a new
   idea: an alignment applied to the cells alone leaves a left heading over a centred control.
   ⚠ The cell holds a *"Settled"* badge instead of a tick on a closed seat, and that centres
   too — which is right, because it occupies the same column for the same reason.

   ⚠ DELIBERATELY `.vz-table`-PREFIXED AND NOT `.vz-table--collect`-PREFIXED: the class has one
   call site today, and scoping an alignment to the one table that uses it would make the second
   caller silently wrong. ⚠ The card override `.vz-table--collect .vz-c-pick` is `0,2,0` and
   sits LATER in the file, so it still wins below 64rem — where it sets `align-self`, not
   `text-align`, and so takes nothing away. */
.vz-table .vz-c-pick { text-align: center; }

/* ⚠ THE 44px TARGET IS THE BASE RULE'S NOW (E6+ item 10), so this only undoes the
   negative margins — a table cell has no gutter to reach into and the neighbouring cell
   is a different control's, not this one's label.
   🔴 **F39 CORRECTED THIS COMMENT AND THEN THE RULE.** It used to end *"the cell is
   `text-align: center` and the margins were symmetric, so removing them changes nothing but
   the overhang"*. ⛔ **The cell was NOT centred** — that was the unprefixed declaration F39
   was filed about, which never reached a browser.
   🔴 **AND FIXING THAT ALONE STILL DID NOT MOVE THE TICK**, which is the trap one level down:
   `.vz-check-input` is `display: grid`, so it is a BLOCK box, and `text-align` only moves
   inline content. Measured with the centring on and this rule still at `margin: 0`: the
   heading centred and the tick stayed **flush left at the padding edge**, which is a worse
   screen than before the row. ▶ `margin-inline: auto` is what centres a block box; the
   `margin-block: 0` half is the original overhang fix, unchanged.
   ⚠ The badge a settled seat renders instead IS inline, so `text-align` centres that one. */
.vz-c-pick .vz-check-input { margin-block: 0; margin-inline: auto; }

@media (max-width: 64rem) {
  /* ⚠ EVERY CELL BELOW NEEDS A NAMED AREA, or it is auto-placed in an order
     nobody chose and still measures ZERO overflow because it fits — B5's blind
     spot, which C1b and C2b both re-checked by reading the computed areas. */
  .vz-table--worklist tr {
    grid-template-areas:
      'group   when'
      'no      seats'
      'amount  amount'
      'actions actions';
  }

  .vz-table--worklist .vz-c-name    { grid-area: group; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--worklist .vz-c-when    { grid-area: when; white-space: normal; justify-self: end; text-align: right; }
  .vz-table--worklist .vz-c-no      { grid-area: no; }
  .vz-table--worklist .vz-c-share   { grid-area: seats; justify-self: end; }
  .vz-table--worklist .vz-c-money   { grid-area: amount; }
  .vz-table--worklist .vz-c-actions { grid-area: actions; text-align: left; }

  /* ⚠ THE BANK-CREDITS PANEL IS THE CLEARING PANEL'S TWIN AND SHARES ITS CARD
     (D11 / `PHASE_D §D11e`). Same six columns in the same order, on the same
     page, read by the same person — so it is listed beside it rather than given
     a second set of rules that could drift. Naming both is deliberate: reusing
     the `--clearing` CLASS would have been cheaper and would have put a bank
     STATEMENT list under a class named for the CLEARING account, which is
     exactly the kind of near-synonym `§D11e` Q2 spent a whole answer on.

     🐛 🔴 AND IT IS NOT CALLED `--statement`, BECAUSE D8 ALREADY IS. The ledger
     account page's table is `.vz-table--statement` — an ACCOUNT statement — and
     its phone card is 400 lines further down this file, so it won on source
     order and dressed this panel's cells in the ledger's grid areas: `when`
     became `posted`, `mode` became `event`, `money` became `debit`. Every cell
     resolved to a NAMED area, so the usual "is any cell unplaced" check came
     back clean and the card was still wrong.
     ⚠ `§D11e` Q2 is why: it spent a whole answer on *"reconciled"* meaning three
     things in this product, and *"statement"* means two. **The second sighting
     of that finding in one row, as a CSS class rather than as a column name.** */
  .vz-table--clearing tr,
  .vz-table--bank-credits tr {
    grid-template-areas:
      'code    when'
      'name    name'
      'mode    amount'
      'actions actions';
  }

  .vz-table--clearing .vz-c-code,
  .vz-table--bank-credits .vz-c-code    { grid-area: code; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--clearing .vz-c-when,
  .vz-table--bank-credits .vz-c-when    { grid-area: when; justify-self: end; }
  .vz-table--clearing .vz-c-name,
  .vz-table--bank-credits .vz-c-name    { grid-area: name; }
  .vz-table--clearing .vz-c-mode,
  .vz-table--bank-credits .vz-c-mode    { grid-area: mode; }
  .vz-table--clearing .vz-c-money,
  .vz-table--bank-credits .vz-c-money   { grid-area: amount; }
  .vz-table--clearing .vz-c-actions,
  .vz-table--bank-credits .vz-c-actions { grid-area: actions; text-align: left; }

  /* The mark is a control that changes stored state, so it gets the tap floor
     the receipts table already gives its own row actions. */
  .vz-table--bank-credits .vz-c-actions .vz-action { min-height: var(--vz-tap-min); }

  /* ⚠ THE TICK STAYS BESIDE THE NAME ON A PHONE. A collection grid where the
     control and the person it belongs to are on different lines is how a
     collector unticks the wrong member with a queue waiting. */
  /* 🔴 `F47` — THE THIRD TRACK IS `auto` AND IT IS SIZED BY THE MEMBER'S NAME, so the
     `amount`/`penalty` row inherited the name's width and the subscription figure landed in
     a different place on every card: **an 89px spread down twenty rows**, and `.vz-c-penalty`
     was exactly as wide as the name above it, row for row, to the pixel.
     ▶ **THE MONEY LINES NOW SPAN**, so each ends on the card's right edge no matter how long
     the name is. The tick still sits beside the person it belongs to, which is what the
     comment above this rule is about and what has not changed. */
  .vz-table--collect tr {
    grid-template-columns: auto minmax(0, 1fr) auto;
    grid-template-areas:
      'pick no      name'
      'pick amount  amount'
      'pick penalty penalty'
      'pick total   total'
      'pick ref     ref';
  }

  /* F140: the reference and the cheque date, side by side under the figures on a phone. */
  .vz-table--collect .vz-c-chq     { grid-area: ref; display: flex; flex-wrap: wrap; gap: var(--vz-space-2); }
  .vz-table--collect .vz-c-chq .vz-control { flex: 1 1 8rem; min-width: 0; }

  .vz-table--collect .vz-c-pick    { grid-area: pick; align-self: center; }
  .vz-table--collect .vz-c-no      { grid-area: no; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--collect .vz-c-name    { grid-area: name; justify-self: end; text-align: right; }
  .vz-table--collect .vz-c-money   { grid-area: amount; }
  .vz-table--collect .vz-c-penalty { grid-area: penalty; }
  .vz-table--collect .vz-c-total   { grid-area: total; }

  .vz-table--collect tfoot tr {
    grid-template-columns: minmax(0, 1fr) auto;
    grid-template-areas: 'label total';
  }

  /* F140: the total row's reference cell is empty and has no place on the card. */
  .vz-table--collect tfoot .vz-c-chq { display: none; }

  .vz-table--collect tfoot th { grid-area: label; }

  .vz-table--runs tr {
    grid-template-areas:
      'when    mode'
      'taken   skipped'
      'amount  amount'
      'actions actions';
  }

  .vz-table--runs .vz-c-when    { grid-area: when; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--runs .vz-c-mode    { grid-area: mode; justify-self: end; }
  /* ⚠ EXPLICIT CLASSES, NOT `:first-of-type` (corrected while measuring C2c).
     `:first-of-type` matches by ELEMENT type — the first `td` among its siblings —
     not by class, so `.vz-c-no:first-of-type` matched NEITHER count cell. Both fell
     through to the unscoped `.vz-c-no { grid-area: no }` above, naming an area this
     template does not define, and were auto-placed in an order nobody chose. It
     measured ZERO overflow because it still fitted, which is exactly B5s blind spot
     and why the pass reads the computed areas rather than the overflow number. */
  .vz-table--runs .vz-c-taken   { grid-area: taken; }
  .vz-table--runs .vz-c-skipped { grid-area: skipped; justify-self: end; }
  .vz-table--runs .vz-c-money   { grid-area: amount; text-align: left; }
  .vz-table--runs .vz-c-actions { grid-area: actions; text-align: left; }
}

/* The two batch counts, tabular so a column of them lines up. */
.vz-c-taken,
.vz-c-skipped {
  font-variant-numeric: tabular-nums lining-nums;
  white-space: nowrap;
}


/* ===========================================================================
   Callouts and key/value panels — RETROACTIVE, and this is a defect fix (C3b)
   ---------------------------------------------------------------------------
   ⚠ `.vz-callout` AND `.vz-kv` HAVE BEEN ON SEVEN SCREENS SINCE C1b AND HAVE
   NEVER EXISTED IN THIS FILE. Found while building C3b's screens by grepping
   for the classes rather than by looking at a page: the markup is semantic, so
   an unstyled `<div>` and `<dl>` still render, still read in order, still pass
   every contrast and overflow measurement, and still announce correctly to a
   screen reader. The responsive pass measures OVERFLOW, TAP TARGETS and
   CONTRAST — none of which a completely unstyled element fails.

   What it cost is the thing those panels exist for: the auction's
   confirm-before-post preview (`FINANCIAL_MODEL §5`) rendered as bare text with
   no border and no ground, indistinguishable from the prose around it — on the
   one panel whose entire job is to be read before somebody posts money. B5's
   blind spot again, one layer up: this time the missing thing was the whole
   rule set rather than one grid area.

   No new colour: `--vz-info`, `--vz-success` and `--vz-warning` are the tokens
   the badges already use.
   =========================================================================== */

/* ⚠ FINANCE'S `.vz-alert`, ADOPTED WHOLE (E14) — THIS IS THE "NOTIFICATION BANNER"
   INCONSISTENCY THE OWNER REPORTED. Until now a callout was a GREY panel with a 3px
   coloured rule down one edge: the same neutral box whatever it meant, with the meaning
   carried by a sliver of colour most readers never register. The finance portal states a
   message as a TINTED PANEL — `--vz-*-tint` ground, `--vz-*-text` ink, a full border in
   the semantic colour — so a refusal reads as a refusal from across the room.

   ⚠ THE TOKENS FOR THIS HAVE BEEN IN `theme.css` SINCE E4 AND NOTHING USED THEM. E4
   brought Finance's four tint/text pairs across as part of adopting the palette and
   recorded that *"a callout is a tinted panel with legible ink rather than coloured text
   on the page ground"* — then left `.vz-callout` painting itself grey. This is that
   sentence finally being true.

   ⚠ AND COLOUR IS STILL NOT THE ONLY SIGNAL (WCAG 1.4.1). The `role` each template sets
   — `status` against `alert` — is what actually reaches a screen reader, the icon is a
   second visual cue, and the text says which it is. All four ink-on-tint pairs measure
   8.2–9.0, so the ink is legible rather than merely coloured. */
.vz-callout {
  display: flex;
  flex-wrap: wrap;
  align-items: flex-start;
  gap: var(--vz-space-2);
  border: 1px solid var(--vz-info);
  border-radius: var(--vz-radius-md);
  background: var(--vz-info-tint);
  color: var(--vz-info-text);
  padding: var(--vz-space-3) var(--vz-space-4);
  margin: 0 0 var(--vz-space-4);
  font-size: var(--vz-fs-sm);
}


/* ===========================================================================
   The callout's icon (E23) — every callout, not the seven that remembered
   ---------------------------------------------------------------------------
   ⚠ THE PRODUCT WAS INCONSISTENT WITH ITSELF. E14 gave the AUTH screens' callouts
   an icon in markup; the other 76 across the app had none. So a refusal on the
   sign-in page and a refusal on the payout screen were different components, which
   is exactly the drift this phase exists to remove.

   ── ⚠ WHY CSS AND NOT MARKUP ──────────────────────────────────────────────
   **Six callouts choose their variant with an inline ternary** — `$suretyOk ?
   '--ok' : '--warn'`, `$isReversed ? '--warn' : '--ok'`. A markup icon would have
   to be chosen by the same condition at each of those, on money screens, and the
   other 77 would each need an edit that a future call site could forget. A rule
   cannot forget.

   ── ⚠ AND THE GLYPHS LIVE HERE ONLY, SO THERE IS STILL ONE DEFINITION ─────
   `partials/icon.php` no longer carries the four alert glyphs: nothing renders
   them as markup any more, and keeping a second copy there is how two drawings of
   one idea drift apart. The data URIs below were GENERATED from that file's paths
   at the moment they moved, so they are the same drawings, not new ones.

   ── ⚠ ABSOLUTELY POSITIONED, WHICH IS THE WHOLE SAFETY ARGUMENT ───────────
   `.vz-callout` is a flex container whose children include 24 multi-paragraph
   bodies across 12 views, and `§E14` records what happened the last time somebody
   changed how those children lay out. A `::before` in the flex FLOW would become a
   flex item and push every one of those paragraphs onto its own row. Out of flow,
   the children's layout is untouched — the only change is the padding that makes
   room, so nothing that renders correctly today can render differently.
   =========================================================================== */
.vz-callout { position: relative; padding-left: calc(var(--vz-space-4) + 26px); }

.vz-callout::before {
  content: '';
  position: absolute;
  left: var(--vz-space-4);
  /* Aligned to the first line's optical centre, not the box top — the same 1px
     nudge `.vz-callout-icon` used when this was markup. */
  top: calc(var(--vz-space-3) + 1px);
  width: 18px;
  height: 18px;
  /* `currentColor` is what makes the icon match its own ink with no rule per
     variant — the property the markup version had, kept. */
  background: currentColor;
  -webkit-mask-image: var(--vz-callout-glyph);
          mask-image: var(--vz-callout-glyph);
  -webkit-mask-repeat: no-repeat;
          mask-repeat: no-repeat;
  -webkit-mask-size: contain;
          mask-size: contain;
}

/* Neutral information — the default callout. */
.vz-callout::before { --vz-callout-glyph: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='%23000' stroke-width='2' stroke-linecap='round' stroke-linejoin='round'%3E%3Ccircle cx='12' cy='12' r='9'/%3E%3Cpath d='M12 9h.01'/%3E%3Cpath d='M11 12h1v4h1'/%3E%3C/svg%3E"); }

/* Something worked. */
.vz-callout--ok::before { --vz-callout-glyph: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='%23000' stroke-width='2' stroke-linecap='round' stroke-linejoin='round'%3E%3Ccircle cx='12' cy='12' r='9'/%3E%3Cpath d='m9 12 2 2 4-4'/%3E%3C/svg%3E"); }

/* A caution. ⚠ A TRIANGLE, NOT A CIRCLE: shape carries the difference for a reader who cannot separate amber from red,
     which is WCAG 1.4.1's whole point. */
.vz-callout--warn::before { --vz-callout-glyph: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='%23000' stroke-width='2' stroke-linecap='round' stroke-linejoin='round'%3E%3Cpath d='M12 4 2.5 20h19z'/%3E%3Cpath d='M12 10v4'/%3E%3Cpath d='M12 17h.01'/%3E%3C/svg%3E"); }

/* A refusal. */
.vz-callout--danger::before { --vz-callout-glyph: url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 24 24' fill='none' stroke='%23000' stroke-width='2' stroke-linecap='round' stroke-linejoin='round'%3E%3Ccircle cx='12' cy='12' r='9'/%3E%3Cpath d='M12 8v4'/%3E%3Cpath d='M12 16h.01'/%3E%3C/svg%3E"); }

/* ⚠ THE MARKUP ICON IS GONE, AND SO IS ITS RULE (E23). The auth screens rendered
   `.vz-callout-icon` as a real `<svg>` child; the `::before` above replaces it
   everywhere, so keeping this would have drawn TWO icons on those seven callouts. */

/* The message column, so a multi-paragraph callout still wraps under its own text
   rather than under the icon. */
.vz-callout-body { flex: 1 1 0; min-width: 0; }

/* 🐛 ⚠ EVERY OTHER DIRECT CHILD TAKES A WHOLE ROW, AND THIS IS NOT A REFINEMENT — IT IS
   THE BUG THIS COMPONENT WOULD OTHERWISE HAVE SHIPPED. Making `.vz-callout` a flex
   container turns its children into COLUMNS, and **24 callouts across 12 views hold two
   to four `<p>` elements** — the payout screen, both settlement screens, the auction
   close, the ledger. Every one of them would have rendered its paragraphs side by side,
   on money screens, and nothing in the suite measures paragraph order.

   ⚠ THE `:not(.vz-callout-icon)` CLAUSE WENT IN E23, BECAUSE THE CLASS DID. The icon is
   a `::before` now — out of flow, so it is not a flex item and needs no exclusion — and
   a selector excluding a class that no longer exists reads as though it still might.
   `.vz-callout-body` stays: seven auth callouts still wrap their text in one. */
.vz-callout > :not(.vz-callout-body) { flex: 1 1 100%; min-width: 0; }

.vz-callout > p,
.vz-callout-body > p { margin: 0 0 var(--vz-space-2); }
.vz-callout > p:last-child,
.vz-callout-body > p:last-child { margin-bottom: 0; }

.vz-callout--ok {
  border-color: var(--vz-success);
  background: var(--vz-success-tint);
  color: var(--vz-success-text);
}

.vz-callout--warn {
  border-color: var(--vz-warning);
  background: var(--vz-warning-tint);
  color: var(--vz-warning-text);
}

.vz-callout-list {
  margin: 0;
  padding-left: var(--vz-space-5);
}

.vz-callout-list li { margin-bottom: var(--vz-space-1); }
.vz-callout-list li:last-child { margin-bottom: 0; }

/* A money figure inside PROSE, as one unbreakable run (C3c).

   🐛 THE BUG THIS EXISTS FOR: the branch-cash warning wrapped as "…take the till to -"
   / "₹1,54,750.00" at 375, orphaning the minus sign so the amount read POSITIVE on the
   next line. Nothing the responsive sweep measures could see it — no overflow, no
   clipping, `scrollWidth === clientWidth` everywhere, contrast fine. A third shape of
   B5's blind spot: the box fits, the ink fits, and the sentence still lies.

   ⚠ `nowrap` IS THE THING THAT SENT 737px THROUGH THE PAGE on the schedule's "moved
   from" line (see above), so it is scoped to a figure of about a dozen characters and
   measured at 375 before shipping: ~95px inside a 307px content box. In a TABLE cell use
   `.vz-c-money`; this is only for a figure sitting in a sentence. */
.vz-figure {
  white-space: nowrap;
  font-variant-numeric: tabular-nums lining-nums;
}
/* A definition list of figures. Two columns on a comfortable width so the
   labels line up and a reader can run down the values; stacked below that,
   because a label column and a rupee figure do not both fit on a phone. The
   values are tabular for the same reason `.vz-c-money` is. */
.vz-kv {
  display: grid;
  grid-template-columns: max-content minmax(0, 1fr);
  gap: var(--vz-space-2) var(--vz-space-4);
  margin: 0;
}

.vz-kv dt {
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

.vz-kv dd {
  margin: 0;
  color: var(--vz-ink);
  font-variant-numeric: tabular-nums lining-nums;
}

/* 🐛 ⚠ A NOTE INSIDE A `.vz-kv` MUST SPAN THE ROW, OR IT SHIFTS EVERY PAIR AFTER IT.
   Reported by the owner as *"the design is completely gone"* on the seat settlement screen,
   and it was neither a missing stylesheet nor a 500.

   `seat-settlement.php` renders an explanatory `<dd class="vz-form-note">` after the
   contributions figure when the seat is PRIZED — valid HTML, since a `<dt>` may carry
   several `<dd>`s. But this is a two-column grid on auto-placement, so that extra child
   **occupies one cell and pushes every following dt/dd across by one**: labels land in the
   value column and values in the label column, for the rest of the list.

   ⚠ AND THE SECOND SYMPTOM IS WHAT MADE IT LOOK LIKE A BROKEN STYLESHEET. Column one is
   `max-content`, sized across everything placed in it — so once a full SENTENCE lands
   there, the column grows to the width of that sentence, the page gains a horizontal
   scrollbar, the navigation is pushed out of view, and the labels squeezed into the
   remainder wrap one word per line.

   ⚠ IT FIRES ONLY ON A PRIZED SEAT, which is why it survived: the fixture the wire tests
   assert this block against reaches it through a path where the branch is not taken, and
   the assertions are about FIGURES rather than about which column they landed in. An
   unpaired grid child is invisible to every check this product has except looking.

   `grid-column: 1 / -1` makes the note take a whole row, so auto-placement resumes at
   column one and the pairs after it stay paired. */
.vz-kv .vz-form-note { grid-column: 1 / -1; }

@media (max-width: 30rem) {
  .vz-kv { grid-template-columns: minmax(0, 1fr); gap: 0; }
  .vz-kv dt { margin-top: var(--vz-space-3); }
  .vz-kv dt:first-of-type { margin-top: 0; }
}


/* ===========================================================================
   Payouts: the prize worklist, the prize screen and the voucher (C3b)
   ---------------------------------------------------------------------------
   Five tables in the vocabulary the other modules already use. No new colour,
   no JavaScript.

   ⚠ EVERY CELL BELOW NEEDS A NAMED AREA, or it is auto-placed in an order
   nobody chose and still measures ZERO overflow because it fits — B5's blind
   spot, re-checked by reading the computed areas rather than the overflow
   number, as C1b, C2b and C2c each did in turn.
   =========================================================================== */

/* A free-text note in a table cell WRAPS. `.vz-c-moved` is `white-space: nowrap`
   and D1c's release reason borrowed it, sending 737px of horizontal overflow
   through a whole page off one 500-character field. These two columns hold a
   refusal reason and an approval note, both VARCHAR(500). */
.vz-c-note {
  white-space: normal;
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
  min-width: 0;
}

@media (max-width: 64rem) {
  /* The worklist is a WORK QUEUE, so the card leads with the group and with what
     the prize is waiting for — the two things somebody scans a hundred rows for. */
  /* 🔴 `F54` — `prize` SHARED ITS ROW WITH A STATUS CHIP, so the figure's right edge moved
     with the length of *"Waiting for a signature"* / *"Approved — not yet paid"*. Measured
     before the fix: right edge **221** against the card's **342**. Same shape as `F47`, and
     the owner's answer there — *"right edge, consistent per card"* — is the rule being
     applied to the rest of its population. */
  .vz-table--prizes tr {
    grid-template-areas:
      "group   when"
      "no      winner"
      "status  status"
      "prize   prize"
      "actions actions";
  }

  .vz-table--prizes .vz-c-name    { grid-area: group; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--prizes .vz-c-when    { grid-area: when; white-space: normal; justify-self: end; text-align: right; }
  .vz-table--prizes .vz-c-no      { grid-area: no; }
  .vz-table--prizes .vz-c-winner  { grid-area: winner; white-space: normal; justify-self: end; text-align: right; }
  .vz-table--prizes .vz-c-money   { grid-area: prize; text-align: left; }
  .vz-table--prizes .vz-c-status  { grid-area: status; white-space: normal; justify-self: end; }
  .vz-table--prizes .vz-c-actions { grid-area: actions; text-align: left; }

  .vz-table--prizes .vz-c-actions .vz-action { min-height: var(--vz-tap-min); }

  .vz-table--vouchers tr {
    grid-template-areas:
      "code    when"
      "name    name"
      "mode    amount"
      "actions actions";
  }

  .vz-table--vouchers .vz-c-code    { grid-area: code; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--vouchers .vz-c-when    { grid-area: when; justify-self: end; }
  .vz-table--vouchers .vz-c-name    { grid-area: name; }
  .vz-table--vouchers .vz-c-mode    { grid-area: mode; white-space: normal; }
  .vz-table--vouchers .vz-c-money   { grid-area: amount; }
  .vz-table--vouchers .vz-c-actions { grid-area: actions; text-align: left; }

  /* Three short columns: the month, its date and one figure. */
  .vz-table--arrears tr {
    grid-template-areas:
      "no     when"
      "amount amount";
  }

  .vz-table--arrears .vz-c-no    { grid-area: no; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--arrears .vz-c-when  { grid-area: when; justify-self: end; }
  .vz-table--arrears .vz-c-money { grid-area: amount; }

  .vz-table--arrears tfoot tr {
    grid-template-columns: minmax(0, 1fr) auto;
    grid-template-areas: "label total";
  }

  .vz-table--arrears tfoot th    { grid-area: label; }
  .vz-table--arrears .vz-c-total { grid-area: total; }

  /* ⚠ THE NOTE GETS ITS OWN FULL-WIDTH ROW on a phone. It is the only cell here
     that can hold a sentence, and sharing a line with a figure would either
     squeeze it to two words a line or push the card wide. */
  .vz-table--trail tr {
    grid-template-areas:
      "status when"
      "name   amount"
      "note   note";
  }

  .vz-table--trail .vz-c-status { grid-area: status; white-space: normal; }
  .vz-table--trail .vz-c-when   { grid-area: when; justify-self: end; }
  .vz-table--trail .vz-c-name   { grid-area: name; }
  .vz-table--trail .vz-c-money  { grid-area: amount; }
  .vz-table--trail .vz-c-note   { grid-area: note; }

  /* 🔴 `F54` — `amount` shared its row with a voucher `code`, whose length is the operator's
     own format string. ⚠ NOT RENDERED BY THE DRIVE FIXTURE — this one is fixed on the grid,
     which is the same grid `--prizes` was measured on, and the entry says so. */
  .vz-table--payouts tr {
    grid-template-areas:
      "when   status"
      "code   code"
      "amount amount"
      "note   note";
  }

  .vz-table--payouts .vz-c-when   { grid-area: when; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--payouts .vz-c-status { grid-area: status; white-space: normal; justify-self: end; }
  .vz-table--payouts .vz-c-money  { grid-area: amount; text-align: left; }
  .vz-table--payouts .vz-c-code   { grid-area: code; justify-self: end; }
  .vz-table--payouts .vz-c-note   { grid-area: note; }
}


/* ===========================================================================
   Settlements: the three-panel worklist, the seat and group pages, the
   settlement itself and its voucher (C4e)
   ---------------------------------------------------------------------------
   Five more tables in the vocabulary the other modules already use. No new
   colour, no JavaScript, and no new component — the voucher, the callouts, the
   `.vz-kv` panels and `.vz-figure` are all C3b's, reused.

   ⚠ EVERY CELL BELOW NEEDS A NAMED AREA, or it is auto-placed in an order
   nobody chose and still measures ZERO overflow because it fits — B5's blind
   spot, re-checked by reading the computed areas rather than the overflow
   number, as C1b, C2b, C2c and C3b each did in turn.

   ⚠ AND TWO MONEY CELLS IN ONE ROW NEED TWO CLASSES. The children tables carry
   an amount AND a net, and `.vz-c-money` used twice would put both in one area
   — C2c's `:first-of-type` defect arriving by a different route. `.vz-c-total`
   is the second one, already in the vocabulary from `.vz-table--arrears`.
   =========================================================================== */

@media (max-width: 64rem) {
  /* The live worklist is a WORK QUEUE, so the card leads with the group and with
     what the settlement is waiting for — the two things somebody scans a
     hundred rows for. */
  /* ⚠ THE `winner` ROW AND ITS PLACEMENT BOTH WENT WITH THE COLUMN (item 63's row): the seat
     is a sub-line inside the group cell now, so this panel renders no `.vz-c-winner` at all.
     ⛔ **Checked rather than assumed** — that cell survives on `.vz-table--holdbacks` further
     down the same screen, which has its own block; a placement left here would have matched
     nothing, which is the mirror of B5's blind spot: not a cell with no area, but an area
     claiming a cell this register does not render.
     ⚠ AND `status` TAKES A ROW OF ITS OWN, because the badge may now wrap to two lines. */
  /* 🔴 `F54` — AND THIS IS THE WORST OF THE FIVE ON PAPER. `amount` shared its row with
     `type`, whose words are *"Member leaving"*, *"Stopped paying"* and **"Squared off at the
     end of the term"** — a range of roughly 14 to 34 characters, so the figure beside it
     could not have landed twice in the same place down a list. ⚠ Not rendered by the drive
     fixture, whose single group has no settlement; fixed on the grid. */
  .vz-table--settlements tr {
    grid-template-areas:
      "group   when"
      "type    type"
      "amount  amount"
      "status  status"
      "actions actions";
  }

  .vz-table--settlements .vz-c-name    { grid-area: group; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--settlements .vz-c-when    { grid-area: when; white-space: normal; justify-self: end; text-align: right; }
  .vz-table--settlements .vz-c-mode    { grid-area: type; white-space: normal; justify-self: end; text-align: right; }
  .vz-table--settlements .vz-c-money   { grid-area: amount; text-align: left; }
  .vz-table--settlements .vz-c-status  { grid-area: status; white-space: normal; justify-self: end; }
  .vz-table--settlements .vz-c-actions { grid-area: actions; text-align: left; }

  .vz-table--settlements .vz-c-actions .vz-action { min-height: var(--vz-tap-min); }

  /* 🔴 `F54` — `held` shared its row with a status. Not rendered by the drive fixture, which
     has no released holdback; fixed on the grid. */
  .vz-table--holdbacks tr {
    grid-template-areas:
      "group   when"
      "winner  winner"
      "status  status"
      "held    held"
      "actions actions";
  }

  .vz-table--holdbacks .vz-c-name    { grid-area: group; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--holdbacks .vz-c-when    { grid-area: when; justify-self: end; }
  .vz-table--holdbacks .vz-c-winner  { grid-area: winner; white-space: normal; }
  .vz-table--holdbacks .vz-c-money   { grid-area: held; text-align: left; }
  .vz-table--holdbacks .vz-c-status  { grid-area: status; white-space: normal; justify-self: end; }
  .vz-table--holdbacks .vz-c-actions { grid-area: actions; text-align: left; }

  .vz-table--holdbacks .vz-c-actions .vz-action,
  .vz-table--holdbacks .vz-c-actions button { min-height: var(--vz-tap-min); }

  /* ⚠ THE COMPONENT NAME GETS ITS OWN FULL-WIDTH ROW. It is a phrase rather than
     a word — "Taken on by whoever fills the seat" — and sharing a line with a
     figure would squeeze it to two words a line or push the card wide. */
  .vz-table--breakdown tr {
    grid-template-areas:
      "component component"
      "effect    no"
      "amount    amount";
  }

  .vz-table--breakdown .vz-c-name  { grid-area: component; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--breakdown .vz-c-mode  { grid-area: effect; white-space: normal; }
  .vz-table--breakdown .vz-c-no    { grid-area: no; justify-self: end; }
  .vz-table--breakdown .vz-c-money { grid-area: amount; text-align: left; }

  .vz-table--children tr {
    grid-template-areas:
      "winner  winner"
      "amount  total"
      "status  actions";
  }

  .vz-table--children .vz-c-winner  { grid-area: winner; white-space: normal; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--children .vz-c-money   { grid-area: amount; text-align: left; }
  .vz-table--children .vz-c-total   { grid-area: total; justify-self: end; }
  .vz-table--children .vz-c-status  { grid-area: status; white-space: normal; }
  .vz-table--children .vz-c-actions { grid-area: actions; text-align: left; justify-self: end; }

  /* ⚠ THE NOTE GETS ITS OWN FULL-WIDTH ROW, `.vz-table--trail`'s reason: it is
     the only cell here that can hold a sentence — a refusal reason or a
     reversal reason, both VARCHAR(500). */
  .vz-table--settlement-history tr {
    grid-template-areas:
      "when   status"
      "type   amount"
      "note   note";
  }

  .vz-table--settlement-history .vz-c-when   { grid-area: when; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--settlement-history .vz-c-status { grid-area: status; white-space: normal; justify-self: end; }
  .vz-table--settlement-history .vz-c-mode   { grid-area: type; white-space: normal; }
  .vz-table--settlement-history .vz-c-money  { grid-area: amount; justify-self: end; }
  .vz-table--settlement-history .vz-c-note   { grid-area: note; }
}

/* ===========================================================================
   Fourteen tables that had no card of their own (2026-11-19)
   ---------------------------------------------------------------------------
   🐛 FOUND DRIVING AT 375. Each rendered a bare `.vz-table`, so below 64rem it
   fell into the generic grid — the MEMBERS grid — and its cells were placed in
   an order nobody chose. Measured, not assumed:
     - a group's Collections: "Billed" and "Collected" drawn on the same spot,
       the figures reading "₹85,0₹0.00";
     - a group's Payouts: prize, arrears taken and held back all three drawn on
       the same spot — `.vz-c-money`'s unscoped area is `value`, and three cells
       shared it;
     - both "Recent sign-in attempts": "Signed in" printed over the address, the
       date a word a line;
     - the merge comparison and the two settings windows: two bare values side
       by side with nothing saying which was which.
   ⚠ None of them overflowed the page, so no width sweep could see one.
   Every cell below has a named area; every value that is not self-describing
   carries its `.vz-cell-label`, which shows only at this width.
   =========================================================================== */

@media (max-width: 64rem) {
  /* Control panel: receivables by operator, TDS awaiting a certificate. */
  .vz-table--aging tr,
  .vz-table--tds-awaited tr { grid-template-areas: "detail total"; }
  .vz-table--aging .vz-c-detail,
  .vz-table--tds-awaited .vz-c-detail { grid-area: detail; }
  .vz-table--aging .vz-c-total,
  .vz-table--tds-awaited .vz-c-total  { grid-area: total; justify-self: end; text-align: right; }

  /* A whole-account export's downloads, Vazhai's view and the owner's. */
  .vz-table--export-log tr {
    grid-template-areas:
      "when   when"
      "status status"
      "ref    qty";
  }
  .vz-table--export-log .vz-c-when   { grid-area: when; white-space: normal; }
  .vz-table--export-log .vz-c-status { grid-area: status; justify-self: start; white-space: normal; }
  .vz-table--export-log .vz-c-ref    { grid-area: ref; }
  .vz-table--export-log .vz-c-qty    { grid-area: qty; justify-self: end; text-align: right; }

  .vz-table--export-log-own tr {
    grid-template-areas:
      "when   when"
      "status status"
      "ref    ref";
  }
  .vz-table--export-log-own .vz-c-when   { grid-area: when; white-space: normal; }
  .vz-table--export-log-own .vz-c-status { grid-area: status; justify-self: start; white-space: normal; }
  .vz-table--export-log-own .vz-c-ref    { grid-area: ref; }

  /* Your own recent sign-ins (both portals): each part on a line of its own — the
     address carries a browser string that no column could hold. */
  .vz-table--signins-own tr {
    grid-template-areas:
      "detail detail"
      "when   when"
      "ref    ref";
  }
  .vz-table--signins-own .vz-c-detail { grid-area: detail; }
  .vz-table--signins-own .vz-c-when   { grid-area: when; white-space: normal; }
  .vz-table--signins-own .vz-c-ref    { grid-area: ref; }

  /* Control-panel settings: how long records are kept, and upload limits. */
  .vz-table--settings-window tr {
    grid-template-areas:
      "name   name"
      "detail mode";
  }
  .vz-table--settings-files tr {
    grid-template-areas:
      "name    name"
      "detail  mode"
      "ceiling ceiling";
  }
  .vz-table--settings-window .vz-c-name,
  .vz-table--settings-files .vz-c-name   { grid-area: name; white-space: normal; }
  .vz-table--settings-window .vz-c-detail,
  .vz-table--settings-files .vz-c-detail { grid-area: detail; }
  .vz-table--settings-window .vz-c-mode,
  .vz-table--settings-files .vz-c-mode   { grid-area: mode; justify-self: end; text-align: right; }
  .vz-table--settings-files .vz-c-ceiling { grid-area: ceiling; }
  /* The values are `.vz-mono`; their labels are words, in the page's own face. */
  .vz-table--settings-window .vz-mono .vz-cell-label,
  .vz-table--settings-files .vz-mono .vz-cell-label { font-family: var(--vz-font-sans); }

  /* Bids called at an auction. */
  .vz-table--bids tr {
    grid-template-areas:
      "detail money"
      "status actions";
  }
  .vz-table--bids .vz-c-detail  { grid-area: detail; }
  .vz-table--bids .vz-c-money   { grid-area: money; justify-self: end; text-align: right; }
  .vz-table--bids .vz-c-status  { grid-area: status; justify-self: start; white-space: normal; }
  .vz-table--bids .vz-c-actions { grid-area: actions; justify-self: end; width: auto; }

  /* A group's months: what was billed and collected side by side, what is
     outstanding on a line of its own. */
  .vz-table--term tr {
    grid-template-areas:
      "no    status"
      "when  detail"
      "money paid"
      "total total";
  }
  .vz-table--term .vz-c-no     { grid-area: no; }
  .vz-table--term .vz-c-status { grid-area: status; justify-self: end; text-align: right; white-space: normal; }
  .vz-table--term .vz-c-when   { grid-area: when; white-space: normal; }
  .vz-table--term .vz-c-detail { grid-area: detail; justify-self: end; text-align: right; }
  .vz-table--term .vz-c-money  { grid-area: money; justify-self: start; text-align: left; }
  .vz-table--term .vz-c-paid   { grid-area: paid; justify-self: end; text-align: right; }
  .vz-table--term .vz-c-total  { grid-area: total; justify-self: start; text-align: left; }

  /* A group's prizes: four amounts in two pairs, so none shares a spot. */
  .vz-table--group-payouts tr {
    grid-template-areas:
      "no      status"
      "detail  detail"
      "money   penalty"
      "amount  total"
      "code    code";
  }
  .vz-table--group-payouts .vz-c-no      { grid-area: no; }
  .vz-table--group-payouts .vz-c-status  { grid-area: status; justify-self: end; text-align: right; white-space: normal; }
  .vz-table--group-payouts .vz-c-detail  { grid-area: detail; }
  .vz-table--group-payouts .vz-c-money   { grid-area: money; justify-self: start; text-align: left; }
  .vz-table--group-payouts .vz-c-penalty { grid-area: penalty; justify-self: end; text-align: right; }
  .vz-table--group-payouts .vz-c-amount  { grid-area: amount; justify-self: start; text-align: left; }
  .vz-table--group-payouts .vz-c-total   { grid-area: total; justify-self: end; text-align: right; }
  .vz-table--group-payouts .vz-c-code    { grid-area: code; }

  /* Merging two member records: the field, then the record kept and the one absorbed. */
  .vz-table--merge tr {
    grid-template-areas:
      "name name"
      "keep absorb";
  }
  .vz-table--merge .vz-c-name   { grid-area: name; }
  .vz-table--merge .vz-c-keep   { grid-area: keep; white-space: normal; overflow-wrap: anywhere; }
  .vz-table--merge .vz-c-absorb { grid-area: absorb; justify-self: end; text-align: right; white-space: normal; overflow-wrap: anywhere; }

  /* Support tickets. */
  .vz-table--support tr {
    grid-template-areas:
      "name   name"
      "code   status"
      "branch when";
  }
  .vz-table--support .vz-c-name   { grid-area: name; white-space: normal; }
  .vz-table--support .vz-c-code   { grid-area: code; }
  .vz-table--support .vz-c-status { grid-area: status; justify-self: end; }
  .vz-table--support .vz-c-branch { grid-area: branch; }
  .vz-table--support .vz-c-when   { grid-area: when; justify-self: end; }

  /* The member portal's months of a group. */
  .vz-table--portal-schedule tr { grid-template-areas: "detail total"; }
  .vz-table--portal-schedule .vz-c-detail { grid-area: detail; }
  .vz-table--portal-schedule .vz-c-total { grid-area: total; justify-self: end; text-align: right; }
}

/* ===========================================================================
   Partners and directors (owner, 2026-11-19)
   ---------------------------------------------------------------------------
   🐛 FOUND DRIVING AT 375 (2026-11-19). The module shipped nine tables as bare
   `.vz-table`, so each fell into the generic card grid — the MEMBERS grid — and
   its cells were auto-placed. A long "For" text ("KCF-001 month 1 prize · voucher
   CBE/PV/000001") squeezed the date to a word a line and the Open button sat on
   top of the "Still owed" label. Nothing overflowed, so no sweep could see it.

   ⚠ EVERY CELL HAS A NAMED AREA (B5's blind spot), the text cells get the full
   width, and the money cells pair up in the two columns with their labels.
   `.vz-c-money` is placed explicitly in each, because its unscoped area is
   `value`, which none of these templates declares.
   =========================================================================== */

@media (max-width: 64rem) {
  .vz-table--funders tr {
    grid-template-areas:
      "name    status"
      "role    money"
      "detail  detail"
      "actions actions";
  }
  .vz-table--funders .vz-c-name    { grid-area: name; white-space: normal; }
  .vz-table--funders .vz-c-status  { grid-area: status; justify-self: end; white-space: normal; }
  .vz-table--funders .vz-c-role    { grid-area: role; }
  .vz-table--funders .vz-c-money   { grid-area: money; justify-self: end; text-align: right; }
  .vz-table--funders .vz-c-detail  { grid-area: detail; }
  .vz-table--funders .vz-c-actions { grid-area: actions; justify-self: start; width: auto; }

  /* A person's entries: what is still owed leads, the purpose has a line of its own. */
  .vz-table--funder-advances tr {
    grid-template-areas:
      "when    total"
      "detail  detail"
      "money   paid"
      "actions actions";
  }
  .vz-table--funder-advances .vz-c-when    { grid-area: when; white-space: normal; }
  .vz-table--funder-advances .vz-c-total   { grid-area: total; justify-self: end; text-align: right; }
  .vz-table--funder-advances .vz-c-detail  { grid-area: detail; }
  .vz-table--funder-advances .vz-c-money   { grid-area: money; justify-self: start; text-align: left; }
  .vz-table--funder-advances .vz-c-paid    { grid-area: paid; justify-self: end; text-align: right; }
  .vz-table--funder-advances .vz-c-actions { grid-area: actions; justify-self: start; width: auto; }

  .vz-table--funder-repayments tr {
    grid-template-areas:
      "when    status"
      "money   mode"
      "ref     ref"
      "actions actions";
  }
  .vz-table--funder-repayments .vz-c-when    { grid-area: when; white-space: normal; }
  .vz-table--funder-repayments .vz-c-status  { grid-area: status; justify-self: end; white-space: normal; }
  .vz-table--funder-repayments .vz-c-money   { grid-area: money; justify-self: start; text-align: left; }
  .vz-table--funder-repayments .vz-c-mode    { grid-area: mode; justify-self: end; }
  .vz-table--funder-repayments .vz-c-ref     { grid-area: ref; }
  .vz-table--funder-repayments .vz-c-actions { grid-area: actions; justify-self: start; width: auto; }

  .vz-table--funder-lines tr {
    grid-template-areas:
      "when  status"
      "money mode"
      "detail detail";
  }
  .vz-table--funder-lines .vz-c-when   { grid-area: when; white-space: normal; }
  .vz-table--funder-lines .vz-c-status { grid-area: status; justify-self: end; white-space: normal; }
  .vz-table--funder-lines .vz-c-money  { grid-area: money; justify-self: start; text-align: left; }
  .vz-table--funder-lines .vz-c-mode   { grid-area: mode; justify-self: end; text-align: right; }
  .vz-table--funder-lines .vz-c-detail { grid-area: detail; }

  .vz-table--funder-story tr {
    grid-template-areas:
      "when when"
      "qty  money";
  }
  .vz-table--funder-story .vz-c-when  { grid-area: when; white-space: normal; }
  .vz-table--funder-story .vz-c-qty   { grid-area: qty; justify-self: start; text-align: left; }
  .vz-table--funder-story .vz-c-money { grid-area: money; justify-self: end; text-align: right; }

  /* Pay back: the box to type in takes the whole width, so a thumb can reach it. */
  .vz-table--funder-repay tr {
    grid-template-areas:
      "name   total"
      "detail detail"
      "when   when"
      "amount amount";
  }
  .vz-table--funder-repay .vz-c-name   { grid-area: name; white-space: normal; }
  .vz-table--funder-repay .vz-c-total  { grid-area: total; justify-self: end; text-align: right; }
  .vz-table--funder-repay .vz-c-detail { grid-area: detail; }
  .vz-table--funder-repay .vz-c-when   { grid-area: when; white-space: normal; }
  .vz-table--funder-repay .vz-c-amount { grid-area: amount; justify-self: stretch; text-align: left; }

  .vz-table--funder-paid tr {
    grid-template-areas:
      "when  code"
      "money money";
  }
  .vz-table--funder-paid .vz-c-when  { grid-area: when; white-space: normal; }
  .vz-table--funder-paid .vz-c-code  { grid-area: code; justify-self: end; }
  .vz-table--funder-paid .vz-c-money { grid-area: money; justify-self: start; text-align: left; }

  .vz-table--funder-cover tr {
    grid-template-areas:
      "name  when"
      "money total";
  }
  .vz-table--funder-cover .vz-c-name  { grid-area: name; white-space: normal; }
  .vz-table--funder-cover .vz-c-when  { grid-area: when; justify-self: end; }
  .vz-table--funder-cover .vz-c-money { grid-area: money; justify-self: start; text-align: left; }
  .vz-table--funder-cover .vz-c-total { grid-area: total; justify-self: end; text-align: right; }

  .vz-table--funder-group tr {
    grid-template-areas:
      "name   when"
      "detail detail"
      "money  total";
  }
  .vz-table--funder-group .vz-c-name   { grid-area: name; white-space: normal; }
  .vz-table--funder-group .vz-c-when   { grid-area: when; justify-self: end; }
  .vz-table--funder-group .vz-c-detail { grid-area: detail; }
  .vz-table--funder-group .vz-c-money  { grid-area: money; justify-self: start; text-align: left; }
  .vz-table--funder-group .vz-c-total  { grid-area: total; justify-self: end; text-align: right; }
}

/* 🐛 ⚠ THE SETTLEMENT HISTORY LINKS ITS *DATE*, AND A DATE IS 17px TALL (C4e).
   Every other table in this product puts its link in `.vz-c-code` or `.vz-c-name`,
   both of which already carry this rule; the settlement history is the first to
   make a `.vz-c-when` cell the link, and it inherited none of it — 73×17 against
   the 24px minimum, at EVERY viewport rather than only on a phone.

   ⚠ FOUND BY MEASURING IN A BROWSER, NOT BY LOOKING, and the green suite could
   not have seen it: the markup is correct, the link works, nothing overflows.
   It is `.vz-c-code a`'s defect exactly — B4 fixed it on a page subtitle, C2b on
   a receipt code, and this is its third sighting. */
.vz-table--settlement-history .vz-c-when a {
  display: inline-flex;
  align-items: center;
  min-height: var(--vz-tap-min);
  color: inherit;
}

.vz-table--settlement-history .vz-c-when a:hover { text-decoration: underline; }

.vz-table--settlement-history .vz-c-when a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}


/* ===========================================================================
   The voucher (C3b)
   ---------------------------------------------------------------------------
   The one page in this product meant to leave the screen on paper, so it is
   bounded and it prints. `receipts.pdf_file_id` exists unwritten and `payouts`
   deliberately has no such column (`§C3`'s scope note) — a browser print is the
   whole of v1's answer, and it therefore has to actually work.
   =========================================================================== */

.vz-voucher {
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  padding: var(--vz-space-5);
  max-width: 46rem;
}

.vz-voucher-declaration {
  margin: var(--vz-space-5) 0 var(--vz-space-6);
  color: var(--vz-ink);
}

/* Two signature blocks side by side, stacked when there is no room for two. */
.vz-voucher-signatures {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-6);
}

.vz-voucher-sign {
  flex: 1 1 14rem;
  /* Room to actually sign, then the rule and its caption underneath. */
  padding-top: var(--vz-space-7);
  box-shadow: inset 0 1px 0 var(--vz-border-strong);
  color: var(--vz-subtle);
  font-size: var(--vz-fs-sm);
  margin: 0;
}

/* ===========================================================================
   The letterhead on a printed document (`§7` item 120)
   ---------------------------------------------------------------------------
   Who the paper is FROM, above the three documents that leave the screen: the
   arrears notice and the two vouchers. It prints — the `@media print` block
   further down hides the CHROME, and this is not chrome.

   ⚠ BOUNDED TO THE SAME 46rem AS `.vz-voucher` so the head and the body share an
   edge; a head wider than the document under it reads as a mis-set page, and on
   paper there is no scrollbar to explain it.
   =========================================================================== */

.vz-letterhead {
  box-sizing: border-box;
  max-width: 46rem;
  margin: 0 0 var(--vz-space-4);
  padding: 0 0 var(--vz-space-3);
  border-bottom: 1px solid var(--vz-border-strong);
}

.vz-letterhead-name {
  margin: 0;
  font-size: var(--vz-fs-md);
  font-weight: var(--vz-fw-bold);
  color: var(--vz-ink);
}

/* The address and the registration numbers: the same size, because they are the
   same kind of fact about the sender and a hierarchy between them would be one
   this product invented. */
.vz-letterhead-line {
  margin: var(--vz-space-1) 0 0;
  font-size: var(--vz-fs-sm);
  color: var(--vz-muted);
}

/* ===========================================================================
   The DRAFT band on an unapproved arrears notice (`§7` item 44)
   ---------------------------------------------------------------------------
   🔴 IT PRINTS, AND THAT IS THE WHOLE POINT. The demand paragraph carries the
   operator's name and cites the Act; until they tick
   `arrears_notice_wording_approved` the wording is a draft. A band that showed on
   screen and vanished on paper would look handled and hand over unapproved legal
   wording anyway — so this one survives the print rules below and makes an
   unapproved notice useless AS a notice, which is the right failure mode for a
   document somebody is given.

   ⚠ NOT `.vz-notice--warn`. That is a screen callout and the print block hides
   `.vz-callout`; this has to be the opposite of hidden.
   =========================================================================== */

.vz-doc-draft {
  margin: 0 0 var(--vz-space-4);
  padding: var(--vz-space-3) var(--vz-space-4);
  border: 2px solid var(--vz-danger);
  border-radius: var(--vz-radius-md);
  color: var(--vz-danger);
  font-weight: var(--vz-fw-bold);
  text-align: center;
  letter-spacing: 0.04em;
  text-transform: uppercase;
  /* ⚠ THE BAND'S EDGE HAS TO LINE UP WITH THE DOCUMENT UNDER IT. `.vz-voucher` is
     `max-width: 46rem` with its own padding, so a content-box band would sit wider
     than the notice by exactly its padding — visible on paper, which is where this
     one is read. */
  box-sizing: border-box;
  max-width: 46rem;
}

/* The Download and Print controls, which sit ABOVE the paper and outside `.vz-voucher`
   so none of the voucher's document styling reaches them (owner, 2026-10-21). */
.vz-voucher-controls {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-gap-sm, 0.5rem);
  margin-bottom: 1rem;
}

/* The download is a one-button form; it must not introduce a block of its own. */
.vz-voucher-controls form { margin: 0; }

@media print {
  /* ⚠ THE CHROME IS NOT THE VOUCHER. Printing the navigation, the branch chip
     and the account menu onto a document a member signs wastes the top third of
     the page and puts a staff member's name on the member's copy.
     🔴 **AND THE DOWNLOAD/PRINT STRIP IS CHROME BY THAT SAME ARGUMENT** — a printed
     voucher carrying a picture of a Print button is the same mistake one control
     along, and it is the one a reader notices because they caused it. */
  /* 🔴 **`--ok` AND NOT EVERY CALLOUT, AND THAT IS A DEFECT THIS FIXED.** The rule
     used to hide `.vz-callout` outright. A transient flash — *"The prize has been
     paid"* — is chrome and must not print. ⛔ But the SETTLEMENT voucher carries its
     reversal notice in `.vz-callout--warn`, so a reversed settlement printed as a
     piece of paper that reads exactly like a live one: no banner, no date, no reason.
     `0056` names that danger in the payout table's own terms.
     ⚠ Measured before narrowing: across the three printable views there is exactly ONE
     `--warn`/`--danger` callout — that reversal notice — and every other is an `--ok`
     flash. So this changes the one element that must print and nothing else. */
  .vz-topbar,
  .vz-nav,
  .vz-callout--ok,
  .vz-voucher-controls,
  .vz-page-sub { display: none; }

  .vz-voucher {
    border: 0;
    padding: 0;
    max-width: none;
  }

  /* 🔴 STATED AGAIN INSIDE THE PRINT BLOCK RATHER THAN LEFT TO INHERIT. The rule
     above already prints, but the border and the colour are exactly what a
     `print-color-adjust` default or a mono printer would drop — and a draft band
     that printed as plain black text reading "DRAFT" in a paragraph is a band
     somebody misses. `!important` is not used: nothing above it competes. */
  .vz-doc-draft {
    border-width: 3px;
    print-color-adjust: exact;
    -webkit-print-color-adjust: exact;
  }
}


/* The maker/checker trail on the voucher is a SEPARATE record from the signature
   blocks above it, and driving C3b showed why that has to be visible: without a
   rule and a heading between them, the "For the branch" caption reads as the
   heading for the trail, so the operator's blank signature line and the names of
   the people who already signed run together as one block. One is a space for a
   person to sign; the other is who did. */
.vz-voucher-trail {
  margin-top: var(--vz-space-6);
  padding-top: var(--vz-space-4);
  border-top: 1px solid var(--vz-border);
}

.vz-voucher-trail .vz-panel-title {
  margin-top: 0;
}


/* ⚠ THE BOX FITTING IS NOT THE TEXT FITTING — B5's blind spot in a new shape (C3b).
   `.vz-kv`'s value column is `minmax(0, 1fr)`, which lets the TRACK shrink below its
   content, so `getBoundingClientRect()` reports a cell inside its card, the page
   reports zero horizontal overflow, and a 44-character ledger transaction id still
   spills out of its own cell and gets clipped. Found on the voucher at 375px by
   measuring `scrollWidth` against `clientWidth` rather than by measuring the box:
   308 vs 293. The clipped field was the LEDGER REFERENCE, which is the one thing on
   that piece of paper somebody reconciling needs in full.

   `overflow-wrap: anywhere` rather than `word-break: break-all`: it only breaks a
   word when there is no other opportunity, so member names and ordinary prose still
   wrap at spaces. */
.vz-kv dd {
  overflow-wrap: anywhere;
}


/* ===========================================================================
   The ledger: the chart of accounts, one account's statement, one transaction
   and both its sides (D8)
   ---------------------------------------------------------------------------
   Three more tables in the vocabulary the other modules already use, one
   callout tone, and two cell classes. No JavaScript, no new component, and
   no new colour token — `--vz-danger` has been in `theme.css` since S5.

   ⚠ EVERY CELL BELOW NEEDS A NAMED AREA, or it is auto-placed in an order
   nobody chose and STILL MEASURES ZERO OVERFLOW because it fits — B5's blind
   spot, re-checked by reading the computed grid areas rather than the overflow
   number, as C1b, C2b, C2c, C3b and C4e each did in turn.
   =========================================================================== */

/* ⚠ AN UNBALANCED TRANSACTION IS NOT A WARNING, IT IS AN INCIDENT
   (`PROJECT_PLAN §9a` Tier 4: "an empty result is the only acceptable answer").
   `--warn` is the tone for "look at this"; this is the tone for "stop". */
/* The refusal. Same shape as its three siblings above (E14) — it lives down here
   because C1b added it in a later block, not because it is a different component. */
.vz-callout--danger {
  border-color: var(--vz-danger);
  background: var(--vz-danger-tint);
  color: var(--vz-danger-text);
}

/* 🐛 ⚠ THE FIELD C3b's DEFECT WAS ON, GIVEN A CLASS OF ITS OWN. A `txn_id` is
   44 characters with no spaces in it, and on the voucher at 375px it spilled
   out of its cell and was clipped while the page reported ZERO overflow —
   `.vz-kv`'s `minmax(0, 1fr)` letting the track shrink below its content. D8
   puts that string on three pages and in a table cell, which is a narrower box
   than the one it already broke, so it carries the fix rather than rediscovering
   it. `overflow-wrap: anywhere` and not `word-break: break-all`, C3b's reason:
   it only breaks a word when there is no other opportunity. */
.vz-c-ref {
  overflow-wrap: anywhere;
  font-variant-numeric: tabular-nums lining-nums;
  font-size: var(--vz-fs-sm);
}

/* The link on that reference is the ONLY way to the transaction page, so it is a
   real target — `.vz-c-code a`'s rule, whose absence C4e found for the third
   time on a cell nobody had made a link before. */
.vz-c-ref:is(a),
.vz-c-ref a {
  display: inline-flex;
  align-items: center;
  min-height: var(--vz-tap-min);
  color: inherit;
}

.vz-c-ref:is(a):hover,
.vz-c-ref a:hover { text-decoration: underline; }

.vz-c-ref:is(a):focus-visible,
.vz-c-ref a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* The running balance reads as a column of figures down the page, so it gets
   `.vz-c-money`'s treatment rather than being a plain cell. */
.vz-c-running {
  /* ⚠ F38 took `text-align: right` out — see the base table section. */
  white-space: nowrap;
  font-variant-numeric: tabular-nums lining-nums;
}

/* ⚠ THE BROUGHT-FORWARD ROW IS NOT AN ENTRY, and must not read as one. It is
   muted and italic so a reader scanning the debit and credit columns does not
   count it as a posting — it is a balance, and it has no amount of its own. */
.vz-row--opening td {
  color: var(--vz-subtle);
  font-style: italic;
}

@media (max-width: 64rem) {
  /* The chart is scanned for ONE thing — what an account holds — so the card
     leads with the account and puts the balance immediately under it, with the
     sentence that says what that balance means below both. */
  /* 🔴 `F47` — `balance` SAT IN THE `1fr` TRACK BESIDE `kind`, AND `kind` IS A WORD THAT
     CHANGES LENGTH ROW BY ROW — "Cash", "Clearing", "Payout holdback". Measured at 375 over
     nine accounts: **96px of spread on the balance's right edge**, the figure sliding left
     as the word beside it grew. ⚠ And the word is a near-duplicate of the account name
     already printed above it, so a balance was being pushed around by a repetition. */
  .vz-table--accounts tr {
    grid-template-areas:
      "account kind"
      "balance balance"
      "means   means"
      "actions actions";
  }

  .vz-table--accounts .vz-c-name    { grid-area: account; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  /* ⚠ `align-self: center` CAME WITH `F47`'s MOVE. The kind used to share a row with the
     balance, where two same-sized cells lined up by themselves; beside the account NAME it
     is a small word against a larger one, and left to stretch it hangs off the top of a
     title that may be two lines deep. */
  .vz-table--accounts .vz-c-mode    { grid-area: kind; white-space: normal; justify-self: end; align-self: center; text-align: right; }
  .vz-table--accounts .vz-c-money   { grid-area: balance; }
  /* ⚠ `white-space: normal` IS LOAD-BEARING HERE. These are sentences —
     "The till has gone overdrawn — more cash has been paid out than was ever
     taken in." — and `.vz-sub-reason`'s defect was exactly this: a cell that
     inherited `nowrap` from a class written for short dates sent 737px of
     horizontal overflow through the whole page. */
  .vz-table--accounts .vz-c-note    { grid-area: means; white-space: normal; }
  .vz-table--accounts .vz-c-actions { grid-area: actions; text-align: left; }

  .vz-table--accounts .vz-c-actions .vz-action { min-height: var(--vz-tap-min); }

  /* A statement card is read as one entry: when, what, how much, and where it
     stands afterwards. ⚠ THE DEBIT AND CREDIT COLUMNS SHARE A ROW and only one
     of them is ever filled, so the empty one is an empty area rather than a gap
     somebody has to interpret. */
  .vz-table--statement tr {
    grid-template-areas:
      "posted  posted"
      "event   event"
      "debit   credit"
      "running running"
      "ref     ref";
  }

  .vz-table--statement .vz-c-when    { grid-area: posted; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--statement .vz-c-mode    { grid-area: event; white-space: normal; }
  .vz-table--statement .vz-c-money   { grid-area: debit; text-align: left; }
  .vz-table--statement .vz-c-total   { grid-area: credit; justify-self: end; }
  .vz-table--statement .vz-c-running { grid-area: running; text-align: left; }
  .vz-table--statement .vz-c-note    { grid-area: ref; }

  /* ⚠ THE `tfoot` ROW IS A GRID TOO, and its `th` spans four columns in the
     table view. Without an area of its own it is auto-placed and the carried-
     forward balance lands somewhere nobody chose — C2c's `.vz-table--dues tfoot`
     defect exactly, one module along. */
  .vz-table--statement tfoot tr {
    grid-template-areas:
      "label   label"
      "running running";
  }

  .vz-table--statement tfoot th { grid-area: label; }

  .vz-table--legs tr {
    grid-template-areas:
      "account account"
      "debit   credit"
      "leg     leg";
  }

  .vz-table--legs .vz-c-name  { grid-area: account; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--legs .vz-c-money { grid-area: debit; text-align: left; }
  .vz-table--legs .vz-c-total { grid-area: credit; justify-self: end; }
  .vz-table--legs .vz-c-no    { grid-area: leg; }

  .vz-table--legs tfoot tr {
    grid-template-areas:
      "label label"
      "debit credit";
  }

  .vz-table--legs tfoot th { grid-area: label; }
}

/* 🐛 ⚠ A LINK INSIDE `.vz-kv` IS 19px TALL, AND D8 IS THE FIRST PAGE TO PUT ONE
   THERE (D8). The transaction page links the group, the seat's member and the
   accounts each leg landed in from inside a `.vz-kv` panel — 19×N against the
   24px minimum, at EVERY viewport rather than only on a phone.

   ⚠ THIS IS THE FOURTH SIGHTING OF ONE DEFECT. `.vz-c-code a` had it (B4, on a
   page subtitle), `.vz-c-name a` had it, C2b found it on a receipt code and C4e
   found it on the settlement history's DATE — every time because a cell that had
   never been a link became one and inherited none of the rule. `.vz-kv dd` is
   that cell this time. The fix is the same three properties, and putting them
   here rather than under a `.vz-table--legs` selector is deliberate: the panel is
   shared, so the next module to link from one gets it for free instead of being
   the fifth sighting.

   ⚠ FOUND BY MEASURING IN A BROWSER, NOT BY LOOKING. The markup is correct, the
   link works, and nothing overflows — the green suite could not have seen it. */
.vz-kv dd a {
  display: inline-flex;
  align-items: center;
  min-height: var(--vz-tap-min);
  color: inherit;
}

.vz-kv dd a:hover { text-decoration: underline; }

.vz-kv dd a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}


/* ===========================================================================
   Reports: the catalogue and the document (D11a)
   ---------------------------------------------------------------------------
   ⚠ THIS IS THE FIRST TABLE IN THE PRODUCT THAT MUST **NOT** BECOME A CARD.

   Every other table carries a `.vz-table--*` modifier naming a grid area per
   cell, because below 44rem `.vz-table` turns each row into a card — and a cell
   with no named area is auto-placed in an order nobody chose while still
   measuring ZERO horizontal overflow, because it fits. That is B5's blind spot,
   and it has now been found five times (`.vz-table--chits`, `--schedule`,
   `--roster`, `--dues tfoot`, `--statement tfoot`).

   A report's columns are NOT KNOWN WHEN THIS FILE IS WRITTEN. They come from a
   `ReportDefinition`: six on a passbook, eight on a day book, four on a trial
   balance, and whatever D11b adds. Authoring a generic card layout for an
   unknown column set would not avoid the auto-placement defect — it would BE
   it, on every report at once and with nobody able to say what the right order
   was. So the honest answer is that a document stays a document: a real table
   that scrolls inside its own wrapper, on every width.

   The page body still never scrolls (`.vz-table-wrap--report` owns the
   overflow), the header row still labels its columns for assistive technology
   instead of being hidden, and nothing is placed anywhere nobody chose.
   =========================================================================== */

.vz-report { margin-top: var(--vz-space-5); }

.vz-report-section { margin-bottom: var(--vz-space-6); }

.vz-report-heading {
  margin: 0 0 var(--vz-space-1);
  font-size: var(--vz-fs-lg);
  color: var(--vz-ink);
}

.vz-report-subtitle {
  margin: 0 0 var(--vz-space-3);
  font-size: var(--vz-fs-md);
  color: var(--vz-ink);
}

.vz-report-sub {
  margin: 0 0 var(--vz-space-3);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

/* ⚠ A DISCLOSURE IS NOT A CAPTION. `§D11` Q1's back-dated line and Q7's missing
   total are part of the document — they go into the CSV too — so they read as
   statements rather than as page furniture, and they WRAP: they are sentences,
   and `.vz-c-moved`'s `nowrap` sent 737px through a whole page once already. */
.vz-report-note {
  margin: 0 0 var(--vz-space-3);
  padding-left: var(--vz-space-3);
  border-left: 2px solid var(--vz-border-strong);
  color: var(--vz-ink);
  font-size: var(--vz-fs-sm);
  white-space: normal;
}

.vz-report-footer {
  margin-top: var(--vz-space-6);
  padding-top: var(--vz-space-4);
  border-top: 1px solid var(--vz-border);
}

.vz-report-footer .vz-form-note { margin-top: var(--vz-space-3); }

/* The report table's own scroll container, so wide documents scroll here and
   never take the page with them. */
.vz-table-wrap--report { overflow-x: auto; }

/* 🐛 ⚠ EVERY LINK IN A REPORT IS A REAL TARGET, AND THIS IS THE FIFTH SIGHTING.
   `.vz-c-code a` established the rule, `.vz-c-ref a` repeated it for C4e's
   transaction reference, `.vz-kv dd a` repeated it again for D8's reference
   panel — and each time it was found by MEASURING, never by looking: the markup
   is correct, the link works, nothing overflows.

   Measured here at 375 on a real day book: **forty-four transaction links at
   296x17px**, every one of them under the 44px minimum. The cause is that a
   report's cells are generic (`.vz-c-note`, `.vz-c-money`) rather than the named
   classes those earlier fixes hang on — the engine does not know that a column
   holds a reference. So the rule goes on ANY link inside a report table, which
   also catches the trial balance's account links for free instead of being the
   sixth sighting. */
/* 🐛 ⚠ AND THE SEVENTH SIGHTING WAS THE SAME CELL AGAIN, THIS TIME AS A WIDTH.
   `.vz-c-ref` carries `overflow-wrap: anywhere` because a 44-character `txn_id` has no
   break opportunity in it; the ledger's own statement gets that class from its view and
   a REPORT cannot, for the reason written directly above — the engine does not know a
   column holds a reference. So the day book asked for a 337px column it could not
   shrink: measured at **1280px desktop**, the table scrolled **121px inside its wrapper**
   while the POSTED column was squeezed to **three lines**, on the register an accountant
   opens every morning. The wrapper was doing its job and the page reported zero overflow.

   The same generalisation, for the same reason: the rule goes on ANY link inside a report
   table. `anywhere` and not `word-break: break-all` is C3b's argument unchanged — it
   breaks a word only when there is no other opportunity, so a member name in a linked
   cell still wraps at its spaces. Measured after: 121px → **0**, the column 337px → 188px,
   the date cell 98px → 61px tall, and the link keeps its 44px tap height. */
/* 🔴 **A LINK IN A TABLE CELL IS THE PRODUCT'S GREEN, NOT THE BROWSER'S BLUE**
   (owner, 2026-10-21: *"can we add our default link style for no columns instead of blue
   link"*, reading the disbursement view's `NO.` column).

   The style is Finance's, from its source rather than from a palette table —
   `vazhai-finance/public/assets/css/app.css:26` is `a { color: var(--vz-primary);
   text-decoration: none; }`. Measured: **43 views put a link inside a `<td>`** and only
   the two report/doc tables had any rule, so all 43 drew browser blue.

   🔴 **BUT THE TOKEN IS `--vz-primary-strong`, NOT `--vz-primary`, AND THE GATE IS WHAT
   SAID SO.** `ThemeContrastTest` measured the accent at **2.39 on `--vz-bg`** against a
   floor of 4.5 — unreadable as body text. This palette SEPARATES the accent from brand
   text for that reason, which Finance's does not have to; `theme.css` line 44 has said
   *"word wants `--vz-primary-strong`"* since E12, and line 80 defines it as BRAND TEXT
   at **5.20**. ▶ Copying one line of another product's CSS was right about the intent
   and wrong about the value.

   ⛔ **THIS MUST STAY ABOVE `.vz-table--report td a`.** The two selectors have IDENTICAL
   specificity — one class, two types — so source order is the only thing deciding them, the
   same trap this file records for the carded `tfoot` at line ~1187. Below it, every
   statutory document's links would turn green; above it, `--report`'s `color: inherit`
   still wins and a report stays black on paper.

   ⚠ The underline is kept for hover and focus rather than dropped: colour alone is not an
   affordance for a reader who cannot distinguish it, and the cell's text sits right beside
   the link. */
.vz-table td a {
  color: var(--vz-primary-strong);
  text-decoration: none;
}

.vz-table td a:hover { text-decoration: underline; }

.vz-table--report td a {
  display: inline-flex;
  align-items: center;
  min-height: var(--vz-tap-min);
  color: inherit;
  overflow-wrap: anywhere;
}

.vz-table--report td a:hover { text-decoration: underline; }

.vz-table--report td a:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* ⚠ THE TOTALS ROW IS NOT A DATA ROW, and must not read as one. A subtotal that
   looked like an account would be counted twice by anybody scanning the column
   — the same failure `.vz-row--opening` exists to prevent one module along. */
.vz-table--report tfoot td {
  border-top: 2px solid var(--vz-border-strong);
  font-weight: 600;
  color: var(--vz-ink);
}

@media (max-width: 64rem) {
  /* Opt back out of the card layout, explicitly rather than by `revert`: the
     base rule sets `display: block` on the table, its body, its rows and its
     cells, and hides `thead` from sight. A document needs all five back.

     🔴 **`--doc` IS THE SAME RULE UNDER AN HONEST NAME, ADDED BY `§7` ITEM 44.**
     The arrears notice needs exactly this behaviour and is not a report, and the
     two ways of giving it that were both worse: putting `--report` on a dunning
     notice reads as a copy-paste error to the next person, and copying these
     twenty lines into a `--notice` block would be two conventions for one
     decision — which is what `§7` item 111 found across seven export preambles.
     ⚠ **SO THE SELECTOR WIDENS AND THE RULE DOES NOT MOVE.** It must stay above
     line ~4340's `.vz-table tfoot { display: block }`… no: it must stay BELOW it,
     which is where it already is, and the generic rule's own comment explains that
     source order is the only thing deciding between them.

     ⛔ **WHAT THIS COSTS IS A SIDEWAYS SCROLL ON A NARROW SCREEN**, because
     `white-space: nowrap` keeps the columns intact. That is the trade the report
     took and it is the right one here for the same reason: this page's job is to
     come out of a printer looking like a table, and a phone card cannot be
     printed as one. */
  .vz-table-wrap--report,
  .vz-table-wrap--doc {
    overflow-x: auto;
    border: 1px solid var(--vz-border);
    border-radius: var(--vz-radius-lg);
    background: var(--vz-surface);
  }

  .vz-table--report,
  .vz-table--doc               { display: table; width: 100%; }
  .vz-table--report thead,
  .vz-table--doc thead         {
    display: table-header-group;
    position: static;
    width: auto;
    height: auto;
    overflow: visible;
    clip: auto;
    clip-path: none;
  }
  .vz-table--report tbody,
  .vz-table--doc tbody         { display: table-row-group; }
  .vz-table--report tfoot,
  .vz-table--doc tfoot         { display: table-footer-group; }
  .vz-table--report tr,
  .vz-table--doc tr            { display: table-row; margin: 0; padding: 0; border: 0; background: none; }
  .vz-table--report th,
  .vz-table--report td,
  .vz-table--doc th,
  .vz-table--doc td            {
    display: table-cell;
    padding: var(--vz-space-2) var(--vz-space-3);
    border-bottom: 1px solid var(--vz-border);
    white-space: nowrap;
  }

  /* The catalogue IS a card, because its three columns are known and fixed. */
  .vz-table--reports tr {
    grid-template-areas:
      "report  report"
      "about   about"
      "actions actions";
  }

  .vz-table--reports .vz-c-name    { grid-area: report; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--reports .vz-c-note    { grid-area: about; white-space: normal; }
  .vz-table--reports .vz-c-actions { grid-area: actions; text-align: left; }
  .vz-table--reports .vz-c-actions .vz-action { min-height: var(--vz-tap-min); }

  /* The filing checklist (D11c). A card is read as one obligation: which form,
     what it is about, and — the reason anybody opened this screen — when it is
     due and whether that date has passed. The deadline and its badge share the
     last row so "30 Apr 2026" and "Past due" are read as one fact.

     ⚠ THE TWO TABLES HAVE SEPARATE MODIFIERS BECAUSE THEIR COLUMNS DIFFER.
     Sharing one would leave three cells in the filed table matching no grid
     area, auto-placed in an order nobody chose, and STILL MEASURING ZERO
     OVERFLOW because they fit — B5's blind spot, which `START_HERE` says to
     catch by reading the computed grid areas rather than the overflow number. */
  .vz-table--compliance tr {
    grid-template-areas:
      "form    form"
      "subject subject"
      "period  period"
      "due     status";
  }

  /* 🐛 ⚠ THE FORM NAME IS THE CARD'S ONLY CONTROL, SO IT CARRIES THE TAP TARGET.
     Every other list in the product pairs a 25px name link with a separate 44px
     "Open" action in a `.vz-c-actions` column — the report catalogue does exactly
     that. This table has no actions column, so measuring at 375 found eleven
     links at **25px** and nothing else to press: the row was unreachable on a
     phone by `UI_CONVENTIONS`' own minimum. Found by measurement, not by eye. */
  .vz-table--compliance .vz-c-name    { grid-area: form; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--compliance .vz-c-name a,
  .vz-table--compliance-filed .vz-c-name a {
    display: block;
    min-height: var(--vz-tap-min);
    padding-block: var(--vz-space-2);
  }
  /* ⚠ `white-space: normal` IS LOAD-BEARING. A subject is
     "DEMO-20M · Demo 1 Lakh — 20 Months · Installment 1", and `.vz-sub-reason`'s
     defect was exactly a cell inheriting `nowrap` from a class written for short
     dates — 737px of horizontal overflow through the whole page. */
  .vz-table--compliance .vz-c-subject { grid-area: subject; white-space: normal; }
  .vz-table--compliance .vz-c-period  { grid-area: period; white-space: normal; }
  .vz-table--compliance .vz-c-due     { grid-area: due; text-align: left; }
  .vz-table--compliance .vz-c-status  { grid-area: status; justify-self: end; text-align: right; }
  /* ⚠ THE GENERATE PREVIEW PUTS A LINK IN THIS CELL — "Already there", pointing at
     the filing that covers the row — and it measured 17px. A badge beside it is
     not a control; this is, so it gets the minimum like every other row action. */
  .vz-table--compliance .vz-c-status a {
    display: inline-flex;
    align-items: center;
    min-height: var(--vz-tap-min);
  }

  /* What has been filed is read for a different thing — the day and the
     Registrar's number — so the card leads with those rather than with a
     deadline that no longer applies. */
  .vz-table--compliance-filed tr {
    grid-template-areas:
      "form    form"
      "subject subject"
      "filed   files"
      "ack     ack";
  }

  .vz-table--compliance-filed .vz-c-name    { grid-area: form; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--compliance-filed .vz-c-subject { grid-area: subject; white-space: normal; }
  .vz-table--compliance-filed .vz-c-filed   { grid-area: filed; text-align: left; }
  .vz-table--compliance-filed .vz-c-files   { grid-area: files; justify-self: end; text-align: right; }
  /* An acknowledgment number is a reference somebody reads back over a phone,
     and "Not yet received" is a sentence — both need to wrap. */
  .vz-table--compliance-filed .vz-c-ack     { grid-area: ack; white-space: normal; }
}

/* ===========================================================================
   Follow-ups: the dunning worklist, the member page and the contact log (D9a)
   ---------------------------------------------------------------------------
   Four tables in the vocabulary the other modules already use. No new colour,
   no new component and NO JAVASCRIPT — the log is a page rather than a hundred
   `<dialog>` elements, which is B4's, B5's, C2b's, C3b's and D1c's move.

   ⚠ EVERY CELL BELOW NEEDS A NAMED AREA, or it is auto-placed in an order
   nobody chose and still measures ZERO overflow because it fits — B5's blind
   spot, which C1b, C2b, C2c and C3b each re-checked by reading the computed
   grid areas rather than the overflow number.
   =========================================================================== */

/* The last-contact cell carries a sentence, two badges and sometimes a consent
   note, all in one column. `.vz-c-note` already wraps; this stops the badges
   from being pushed onto a line of their own on a wide screen while keeping
   them readable when they do wrap. */
.vz-table--dunning .vz-c-note .vz-badge { margin-top: var(--vz-space-1); }

/* ⚠ THE STATEMENT MARK STACKS UNDER THE CLEARING BADGE, NEVER BESIDE IT
   (D11 / `PHASE_D §D11e`). A payment row can now carry two badges — the bank
   saying the instrument realised, and a person saying they found it on the
   statement — and side by side they take this column from *"With the bank"* at
   95px to about 200px, on a six-column table that already has to fit. They are
   a state and a footnote to it, so they read as a stack.
   `.vz-badge` is `inline-flex` and `width: max-content` keeps the pill hugging
   its text once it is a block. */
.vz-table--receipts .vz-c-status .vz-badge { display: flex; width: max-content; }
.vz-table--receipts .vz-c-status .vz-badge + .vz-badge { margin-top: var(--vz-space-1); }

/* ⚠ AND THE DATE UNDER IT WRAPS, WHICH `.vz-c-moved` DOES NOT DO BY DEFAULT. It is
   `nowrap` because it usually carries a code beside a name; here it is a short sentence,
   and its one long line set this column's min-content to 203px against the badge's own
   169px. Measured at 1024: 815px of table in a 727px wrap becomes 781px. */
.vz-table--receipts .vz-c-status .vz-c-moved { white-space: normal; }

@media (max-width: 64rem) {
  /* ⚠ THE CARD LEADS WITH THE PERSON AND WHAT THEY OWE, because those are the
     two things somebody scans a chasing list for — and the bucket sits beside
     the name rather than under it, so "who" and "how bad" are one glance. */
  /* ⚠ THE `since` ROW IS GONE WITH THE COLUMN (item 63's row). The oldest due date is now a
     sub-line inside the arrears chip's own cell, so there is no `.vz-c-when` here to place —
     and an area named with no cell claims this register renders something it does not, while
     a CELL with no area is B5's blind spot. **Both halves were removed together.** */
  .vz-table--dunning tr {
    grid-template-areas:
      'member  bucket'
      'amount  amount'
      'contact contact'
      'who     actions';
  }

  .vz-table--dunning .vz-c-name    { grid-area: member; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--dunning .vz-c-status  { grid-area: bucket; justify-self: end; text-align: right; white-space: normal; }
  .vz-table--dunning .vz-c-money   { grid-area: amount; text-align: left; }
  .vz-table--dunning .vz-c-note    { grid-area: contact; white-space: normal; }
  .vz-table--dunning .vz-c-mode    { grid-area: who; }
  .vz-table--dunning .vz-c-actions { grid-area: actions; justify-self: end; text-align: right; }
  /* ⚠ THE SEAT COUNT IS DROPPED ON A PHONE, DELIBERATELY AND NOT BY ACCIDENT.
     It is the one column that is derivable from the member's own page, and a
     card with eight facts on it is one nobody reads. Hidden with an explicit
     rule rather than left unplaced, because an unplaced cell is B5's blind
     spot — it would still render, in a position nobody chose. */
  .vz-table--dunning .vz-c-share   { display: none; }

  .vz-table--callbacks tr {
    grid-template-areas:
      'when    when'
      'member  member'
      'outcome actions'
      'note    note';
  }

  .vz-table--callbacks .vz-c-when    { grid-area: when; white-space: normal; }
  .vz-table--callbacks .vz-c-name    { grid-area: member; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--callbacks .vz-c-mode    { grid-area: outcome; }
  .vz-table--callbacks .vz-c-actions { grid-area: actions; justify-self: end; text-align: right; }
  .vz-table--callbacks .vz-c-note    { grid-area: note; white-space: normal; }

  /* 🐛🔴 RENAMED FROM `.vz-table--seats` BY ITEM 64's HALF OF ITEM 63's ROW, AND THE
     DUPLICATION WAS THE DEFECT.
     ⏸ **TWO DIFFERENT TABLES ON TWO DIFFERENT SCREENS SHARED ONE MODIFIER.** The member
     profile's seats panel renders `name · no · share · status · standing · actions` and has its
     own block ~1,700 lines above this one, correctly placing all six. This block was written
     for the FOLLOW-UP screen's seats panel — `name · no · status · money · actions` — and named
     the same class, so it won on source order **everywhere**: on the member profile
     `.vz-c-share` and `.vz-c-standing` were left with no `grid-area` at all.
     🔴 **MEASURED AT 320: `grid-template-columns` computed to `0px 82.47px 0px 78.53px`** —
     four tracks where the template names two, **two of them zero, each holding a cell with real
     text** ("Settlement Wire Group", "Whole ticket") which spilled across its neighbour. Page
     overflow read 0, the wrap read 0, every test was green. **That is `§7` item 64 exactly**,
     and its stated per-table fix (a one-column card) would have treated the symptom.
     ▶ **The cause was two tables sharing a name**, which `§8a` row 7i warned about from the
     other direction — *"reusing a name another module already styles wins on source order"*.
     ⚠ The block above already carries `§E6d`'s warning that an unplaced cell measures clean;
     this is that same defect arriving through a COLLISION rather than an omission. */
  /* 🔴 `F54` — `amount` shared its row with the ACTIONS cluster, which is the most variable
     neighbour of the five: `.vz-c-actions` is a wrapping flex row whose width is however many
     controls this seat happens to offer. Not rendered by the drive fixture, which seeds no
     follow-up; fixed on the grid. */
  .vz-table--followup-seats tr {
    grid-template-areas:
      'group   group'
      'seat    bucket'
      'amount  amount'
      'actions actions';
  }

  .vz-table--followup-seats .vz-c-name    { grid-area: group; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--followup-seats .vz-c-no      { grid-area: seat; }
  .vz-table--followup-seats .vz-c-status  { grid-area: bucket; justify-self: end; text-align: right; white-space: normal; }
  .vz-table--followup-seats .vz-c-money   { grid-area: amount; text-align: left; }
  .vz-table--followup-seats .vz-c-actions { grid-area: actions; justify-self: end; text-align: right; }

  /* The contact history is READ rather than worked, so the card leads with when
     and who and lets the note run the full width underneath. */
  .vz-table--history tr {
    grid-template-areas:
      'logged  agent'
      'channel outcome'
      'promise promise'
      'note    note';
  }

  .vz-table--history .vz-c-when   { grid-area: logged; white-space: normal; }
  .vz-table--history .vz-c-name   { grid-area: agent; justify-self: end; text-align: right; white-space: normal; }
  .vz-table--history .vz-c-mode   { grid-area: channel; }
  .vz-table--history .vz-c-status { grid-area: outcome; justify-self: end; text-align: right; white-space: normal; }
  .vz-table--history .vz-c-money  { grid-area: promise; text-align: left; white-space: normal; }
  .vz-table--history .vz-c-note   { grid-area: note; white-space: normal; }
}

/* ⚠ PRINT IS THE PDF, AND `§D11` Q4 SAYS SO ON PURPOSE. `composer.json` has no
   runtime dependencies and a server PDF needs one, so `payouts/voucher.php`'s
   precedent stands: this is a page, and the browser prints it. What print must
   not carry is the app's chrome or the controls that produced the document. */
@media print {
  .vz-toolbar,
  .vz-page-actions,
  .vz-report-footer .vz-form-note + * { display: none; }

  .vz-table-wrap--report { overflow: visible; border: 0; }
  .vz-report-section     { break-inside: auto; }
  .vz-report-heading     { break-after: avoid; }
}


/* ===========================================================================
   THE MEMBER'S PASSBOOK — `OPERATING_MODE §7` ITEM 92 (PHASE_O §O1d)
   ---------------------------------------------------------------------------
   🐛 MEASURED 2026-09-16 BUILDING §O1c: the member's passbook rendered
   `.vz-table--report` and scrolled sideways **552px inside a 299px wrap at 375
   (253 over)** and **552 in 244 at 320**, with *Balance after* — the column they
   came for — the rightmost one off-screen. ⚠ `documentElement.scrollWidth -
   clientWidth` read **0 at every width 320→1440**, because
   `.vz-table-wrap--report` is a proper scroll container doing its job. The
   ordinary sweep certifies the page and says nothing about the table.

   ⛔ THE FIX IS NOT A RULE CHANGE ON `--report`, AND ITEM 92 SAYS SO. That
   modifier's opt-out is deliberate — *"a document needs all five back"* — it is
   right for an operator reading a register at a desk, and every report screen in
   the product depends on it. The attribution was controlled both ways: removing
   the class from the captured page carded the table at 299px, so the scroll is
   the modifier's and not the portal's markup.

   ▶ SO THE MEMBER'S COPY WEARS ITS OWN MODIFIER AND CARDS LIKE EVERY OTHER TABLE
   IN THE PRODUCT. `.vz-table--reports` (the catalogue) is the precedent — *"the
   catalogue IS a card, because its three columns are known and fixed"* — and the
   member's passbook has five known, fixed columns. It is NOT a statutory
   document: the operator's own `member_passbook` at `/app/reports` still prints
   through `--report`, untouched, from the same derivation.

   🔴 EVERY CELL CLASS IS SCOPED TO THIS MODIFIER, AND THAT IS NOT TIDINESS.
   `.vz-c-money { grid-area: value }` and `.vz-c-when { grid-area: when }` are
   UNSCOPED rules inside the card block above, belonging to `--chits` and
   `--auctions`. `§8a` row 7i records what inheriting another table's areas costs:
   every cell resolves to a *named* area — somebody else's — so the "every td has
   a grid-area" check passes while the card is wrong, at 0px overflow.
   =========================================================================== */

/* ⚠ THE TOTALS ROW IS NOT A DATA ROW. `.vz-table--report tfoot td` carries the
   identical rule for the same reason: a subtotal that looked like a movement
   would be counted twice by anybody scanning the column. */
.vz-table--passbook tfoot td {
  border-top: 2px solid var(--vz-border-strong);
  font-weight: var(--vz-fw-bold);
  color: var(--vz-ink);
}

/* 🔴 THE CARD'S OWN COLUMN LABELS, AND THEY EXIST BECAUSE THE `thead` STOPS
   WORKING. Below the breakpoint the table is `display: block`, which drops table
   semantics from the accessibility tree — so the screen-reader-only `thead` is
   labelling nothing, and a sighted reader sees two rupee figures side by side
   with no nouns. `§E6` row 14 recorded exactly that: *"a value with no noun on a
   phone card that has no column headings."*

   ⚠ `display: none` ABOVE THE BREAKPOINT, NOT `.vz-sr-only`. Up there the real
   `<th>` does the job, and a hidden duplicate would make a screen reader announce
   every column twice — `passbook.php`'s `<caption>` defect, in a second costume. */
.vz-cell-label { display: none; }

@media (max-width: 64rem) {
  .vz-cell-label {
    display: block;
    color: var(--vz-subtle);
    font-size: var(--vz-fs-xs);
    font-weight: var(--vz-fw-bold);
    letter-spacing: 0.06em;
    text-transform: uppercase;
  }

  /* 🔴🔴 `F46` + `F47` — A LABELLED MONEY LINE, AND IT FIXES TWO DEFECTS WITH ONE BOX.
     The mechanism above shipped in `§E6` row 14 and then served **exactly one view in the
     product** — `portal/passbook.php` — while fifteen operator tables rendered **231 bare
     figures with no noun anywhere on the card**. Owner, 2026-09-25, on the money columns:
     *"money columns only, now"*, and on the alignment: *"right edge, consistent per card"*.

     🔴 **THE TWO ANSWERS ARE ONE RULE, BECAUSE THE SECOND DEFECT IS A GRID DEFECT.**
     `.vz-c-money { justify-self: end }` — written for the `--chits` card two thousand lines
     up — leaks into every card that re-homes the cell with `grid-area` and does not re-home
     the alignment. The figure then shrink-wraps to the **end of its own track**, and the
     generic card grid is `minmax(0, 1fr) auto`: **a cell in the non-final column has a right
     edge set by its neighbour's content.** Measured at 375 on 2026-09-25:

         --collect  .vz-c-money   20 rows   right edge spread  89px   (tracks the member name, byte for byte)
         --accounts .vz-c-money    9 rows   right edge spread  96px   (tracks the account kind)
         --schedule .vz-c-money   20 rows   right edge spread   7px

     ▶ **SO THE MONEY CELL TAKES THE WHOLE CARD WIDTH** — see each table's own areas below —
     **and lays its label and its figure out as a line.** The figure then ends on the CARD's
     right edge, which no neighbour can move, and the noun that says what it is sits beside
     it. `--collect`'s three figures were `₹4,250.00 ₹365.50` and a bold `₹4,615.50` with
     nothing naming any of them, on the screen a collector uses at a counter.

     🐛 ⚠ **AND THE TWO WORDS DO DIFFERENT JOBS — MEASURED, BECAUSE THE FIRST DRAFT OF THIS
     COMMENT GOT IT WRONG.** It claimed `justify-self: stretch` was what fixed the right edge.
     Removing it on purpose and re-measuring says otherwise: **the spread stays 0 and the
     right edge stays on the card**, because the `grid-area` span above is what does that.
     What `stretch` actually holds is the LABEL. On `/app/chits/{id}/collections/{no}` at 375,
     with it the cell runs `89 → 327` and the noun starts at the card's own text edge, `89`;
     without it the cell shrink-wraps to `169 → 327` and the noun floats in the middle of the
     card, jammed against the figure it names. ▶ **The span fixes the figure; the stretch
     fixes the label.**

     ⚠ **`text-align: right` IS THE THIRD WORD AND IT IS NOT DECORATION EITHER.** Both
     `--collect` and `--accounts` carried `text-align: left` here and it moved nothing at all,
     because a box exactly as wide as its own text has no room to align inside — `F39`'s
     lesson, which `--schedule`'s own comment states correctly and which these two were
     written in ignorance of. Those three dead declarations are deleted.

     ⛔ **SCOPED TO THE TABLES THAT WERE MEASURED**, and not applied to `.vz-c-money`
     generally. `--worklist` and `--arrears` already span both columns and already end on
     the card's right edge; they take the LABEL and nothing else. `portal/passbook.php` is
     untouched — its labels stack above figures that share a row two-up, which is right
     there and would be wrong here.

     ⛔ **`tbody` AND NOT THE WHOLE TABLE, FOR THE SAME REASON `§7` ITEM 93 SCOPED ITS OWN
     ONE-WORD FIX THERE.** `--collect`, `--dues` and `--arrears` each carry a `tfoot` totals
     row with its own two-cell grid and a real `<th>` already doing the labelling. A figure
     there has no `.vz-cell-label`, so it would be the lone item in a `space-between` box —
     which puts it at flex-start, **swinging the closing total of three money screens to the
     left edge**. A totals row is not a card. */
  .vz-table--collect tbody .vz-c-money,
  .vz-table--collect tbody .vz-c-penalty,
  .vz-table--collect tbody .vz-c-total,
  .vz-table--dues tbody .vz-c-money,
  .vz-table--dues tbody .vz-c-penalty,
  .vz-table--dues tbody .vz-c-total,
  .vz-table--schedule tbody .vz-c-money,
  .vz-table--schedule tbody .vz-c-total,
  .vz-table--accounts tbody .vz-c-money,
  .vz-table--worklist tbody .vz-c-money,
  .vz-table--arrears tbody .vz-c-money,
  /* 🔴 `F54` — the rest of the population `F47` measured, found from the grids rather than
     from renders: five more money cells that shared a row with a neighbour of varying length
     (a status chip, a voucher code, a settlement type, an actions cluster). The five that are
     NOT here — `--statement`, `--legs`, `--passbook`, `--children`, `--subscription` — pair
     money WITH money on purpose, and a pair sharing a row is the design rather than a defect. */
  .vz-table--prizes tbody .vz-c-money,
  .vz-table--payouts tbody .vz-c-money,
  .vz-table--settlements tbody .vz-c-money,
  .vz-table--holdbacks tbody .vz-c-money,
  .vz-table--followup-seats tbody .vz-c-money {
    justify-self: stretch;
    display: flex;
    align-items: baseline;
    justify-content: space-between;
    gap: var(--vz-space-3);
    text-align: right;
  }

  /* ⚠ THE LABEL DOES NOT SHRINK THE FIGURE. A long noun — "Outstanding after this" — would
     otherwise squeeze a rupee amount into a wrap, and a money figure that breaks across two
     lines is the one thing this rule exists to prevent. */
  .vz-table--collect .vz-cell-label,
  .vz-table--dues .vz-cell-label,
  .vz-table--schedule .vz-cell-label,
  .vz-table--accounts .vz-cell-label,
  .vz-table--worklist .vz-cell-label,
  .vz-table--arrears .vz-cell-label,
  .vz-table--prizes .vz-cell-label,
  .vz-table--payouts .vz-cell-label,
  .vz-table--settlements .vz-cell-label,
  .vz-table--holdbacks .vz-cell-label,
  .vz-table--followup-seats .vz-cell-label {
    align-self: center;
    min-width: 0;
    overflow-wrap: anywhere;
  }

  /* The date and what happened take the full width; the two money columns share a
     row, because "in" and "out" are read as a pair; the balance takes its own,
     because it is the figure this screen exists for and it should not have to
     compete for the eye with the movement that produced it. */
  .vz-table--passbook tr {
    grid-template-areas:
      "posted  installment"
      "event   event"
      "debit   credit"
      "running running";
  }

  /* 🔴 THE INSTALLMENT SHARES THE DATE'S ROW (owner, 2026-09-28, point 4). Both answer
     *"when"*, and on a card the pair reads as one fact — *"28 Sep · installment 3"* — rather
     than as two lines competing with the movement below them.
     ⛔ IT NEEDED AN AREA OF ITS OWN OR IT WOULD BE AUTO-PLACED, which is this block's own
     recorded defect: the *"every td has a grid-area"* check passes while the cell lands in
     somebody else's, at 0px overflow. */
  .vz-table--passbook .vz-c-list    { grid-area: installment; justify-self: end; color: var(--vz-muted); }
  .vz-table--passbook .vz-c-when    { grid-area: posted; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--passbook .vz-c-mode    { grid-area: event; white-space: normal; }
  .vz-table--passbook .vz-c-money   { grid-area: debit; justify-self: start; text-align: left; }
  .vz-table--passbook .vz-c-total   { grid-area: credit; justify-self: end; text-align: left; }
  .vz-table--passbook .vz-c-running { grid-area: running; text-align: left; }

  /* ⚠ THE `tfoot` ROW IS A GRID TOO, and it has no date — the totals line is not
     dated. Without a template of its own the cells are auto-placed and the closing
     balance lands somewhere nobody chose, which is `.vz-table--statement`'s
     recorded finding and C2c's `--dues` defect before it. */
  .vz-table--passbook tfoot tr {
    grid-template-areas:
      "event   event"
      "debit   credit"
      "running running";
  }

  /* 🐛 F116 — THE MEMBER'S RECEIPTS WERE A BARE `.vz-table`, so on a phone they took the
     MEMBERS grid above, whose areas name none of these cells: the receipt number landed on
     its own date and "RECEIPT" printed over "PAID ON" (2026-11-03 drive, at 375px, 0px
     overflow — it fitted, overlapping). The generic block says a new table must bring its
     own modifier; this is that modifier, and every cell below has an area of its own.
     The amount spans the card because it is the figure on the paper; the two parts it
     splits into (F112) share the row under it, read as a pair. */
  .vz-table--receipts tr {
    grid-template-areas:
      "receipt get"
      "paid    get"
      "amount  amount"
      "dues    fee";
  }

  .vz-table--receipts .vz-c-ref     { grid-area: receipt; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--receipts .vz-c-when    { grid-area: paid; white-space: normal; }
  .vz-table--receipts .vz-c-actions { grid-area: get; align-self: start; justify-self: end; }
  .vz-table--receipts .vz-c-money   { grid-area: amount; justify-self: end; }
  .vz-table--receipts .vz-c-paid    { grid-area: dues; justify-self: start; text-align: left; }
  .vz-table--receipts .vz-c-penalty { grid-area: fee; justify-self: end; }

  /* ✅ `.vz-table--passbook tfoot { display: block }` STOOD HERE AND IS GONE (`§7` item 93,
     2026-09-16). `§O1d` fixed its own table and filed the general rule, because taking one
     word in the generic block without capturing the other six views is what `§7` item 83
     refuses. **Item 93 then captured all six and took it**, so `.vz-table tfoot` in the
     generic block now cards this footer along with everybody else's.
     ⛔ Do not put a scoped copy back: two rules for one fact is how they come to disagree. */

  /* ⛔ `display: none` RATHER THAN AN AREA, because there is nothing in it. An
     empty cell placed in the card draws its own padding and reads as a missing
     value on the one screen where a blank beside money is alarming. */
  .vz-table--passbook tfoot .vz-c-when { display: none; }

  /* ⚠ AND NEITHER IS THE TOTALS LINE AN INSTALLMENT. The `tfoot` template above has no
     `installment` area, so this cell would be auto-placed — the same defect `.vz-c-when`
     is hidden here to avoid. It is empty either way; `display: none` keeps it from
     claiming a grid slot. */
  .vz-table--passbook tfoot .vz-c-list { display: none; }
}

/* --- Colour swatches (E4b) ------------------------------------------------ */

/* ⚠ THE SWATCH IS A CLASS, NOT A `style="background:#…"` ATTRIBUTE, and that is the
   whole point rather than a nicety. `PHASE_E §E8` requires the org's brand colour to be
   served as a CSS ROUTE precisely because an inline block or a style attribute forces
   `style-src 'unsafe-inline'` on any future CSP — and a settings screen that PREVIEWED
   the colour inline would have handed that concession straight back at the one URL where
   the colour is chosen.

   These rules paint `var(--vz-primary…)`, so `/app/branding.css` moves them by moving the
   tokens. The swatch therefore shows what the product is ACTUALLY painting rather than
   what the row says it should be — which is the difference between a preview and a
   measurement, and this row's whole argument is that a screen should show the second.

   ⚠ AND EACH ONE IS `aria-hidden` IN THE VIEW, with the hex written beside it as text.
   A colour is not information a screen reader can convey, and a swatch with an accessible
   name would announce "green" to somebody for whom that is the one useless fact. */
.vz-swatch {
  display: inline-block;
  width: 1.5rem;
  height: 1.5rem;
  vertical-align: -0.35rem;
  margin-right: var(--vz-space-2);
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-sm);
}

.vz-swatch--primary { background: var(--vz-primary); }
.vz-swatch--strong  { background: var(--vz-primary-strong); }
.vz-swatch--hover   { background: var(--vz-primary-hover); }

/* --- E25: five classes that were rendered and painted nothing -------------- */

/* ⚠ THE AUDIT'S FIRST FINDING, AND IT IS THIS REPO'S OLDEST DEFECT FAMILY. Ten classes
   were rendered by views with no rule anywhere: `.vz-c-*` shipped that way in C1b,
   `.vz-table--compliance` in D11c, the whole `.vz-callout` component in C3b,
   `.vz-combo-search` in E18 and `.vz-panel-subtitle` in E26. Five were renames onto
   components that already existed — `.vz-actions` was `.vz-page-actions`, `.vz-detail`
   was `.vz-kv`, `.vz-pill` was `.vz-badge`. These five are the ones where the class was
   right and the rule was simply missing. */

/* An empty state's heading read exactly like its body text, so the state had a title
   that was not one. `.vz-empty` centres the box; this is what makes the first line lead. */
.vz-empty-title {
  margin: 0 0 var(--vz-space-2);
  font-weight: var(--vz-fw-medium);
  color: var(--vz-ink);
}

/* "Powered by" beside the Vazhai mark. Without this the gap between the words and the
   mark was a whitespace text node — which is a space that changes with the font. */
.vz-footer-brand {
  display: inline-flex;
  align-items: center;
  gap: var(--vz-space-2);
}

/* A list of records — the compliance acknowledgements today. Indented like the callout's
   list rather than at the browser's default 40px, which sat it oddly under its heading. */
.vz-list {
  margin: 0 0 var(--vz-space-3);
  padding-left: var(--vz-space-5);
  font-size: var(--vz-fs-sm);
}

.vz-list li { margin-bottom: var(--vz-space-1); }
.vz-list li:last-child { margin-bottom: 0; }

/* ⚠ THE ONE UTILITY CLASS IN THIS STYLESHEET, AND IT IS HERE BECAUSE THE MARKUP ALREADY
   ASSUMED IT. `auctions/index.php` renders a group's code in `.vz-muted` beside its name;
   with no rule the code was full ink and read as a second, equal piece of information. A
   token existed and the class to apply it did not. */
.vz-muted { color: var(--vz-muted); }

/* The wizard's step label, beside `.vz-step-no` which has had a rule all along. */
.vz-step-name { font-size: var(--vz-fs-sm); }

/* --- The brand-colour presets (E26) --------------------------------------- */

/* ⚠ THE COLOURS ARE HERE AND THE HEXES ARE IN `App\Support\BrandPresets`, AND A TEST
   HOLDS THE TWO TOGETHER. A swatch cannot carry `style="background:#2563EB"` — E16 took
   `app/Views/` to zero inline `style` attributes, and a CSP's `style-src` blocks an
   attribute exactly as it blocks a `<style>` element. `BrandPresetTest` asserts every
   class here paints the hex that class's button submits, because a colour that shows one
   value and applies another is the worst possible defect for this control. */
.vz-preset-grid {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2);
  margin-bottom: var(--vz-space-4);
}

/* 44px, because this is the control the row exists for and it is used one-handed on a
   phone as often as at a desk. */
.vz-preset {
  width: var(--vz-tap-min);
  height: var(--vz-tap-min);
  padding: 0;
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  cursor: pointer;
  transition: box-shadow var(--vz-transition), border-color var(--vz-transition);
}

.vz-preset:hover { border-color: var(--vz-ink); }
.vz-preset:focus-visible { outline: 2px solid var(--vz-primary); outline-offset: 2px; }

/* ⚠ THE COLOUR IN USE IS MARKED BY A RING AND A TICK, NEVER BY COLOUR ALONE — which
   would be absurd here in particular: on a grid of ten colours, "the selected one is
   coloured differently" says nothing at all. */
.vz-preset[aria-current='true'] {
  border-color: var(--vz-ink);
  box-shadow: 0 0 0 2px var(--vz-surface), 0 0 0 4px var(--vz-ink);
}

.vz-preset--vazhai  { background: #5DB52E; }
.vz-preset--forest  { background: #1B7F4C; }
.vz-preset--teal    { background: #0E7C86; }
.vz-preset--blue    { background: #2563EB; }
.vz-preset--indigo  { background: #4F46E5; }
.vz-preset--violet  { background: #7C3AED; }
.vz-preset--magenta { background: #BE185D; }
.vz-preset--red     { background: #DC2626; }
.vz-preset--orange  { background: #EA580C; }
.vz-preset--slate   { background: #475569; }

/* The custom row: the native wheel beside the hex box. */
.vz-colour-row {
  display: flex;
  align-items: flex-end;
  gap: var(--vz-space-3);
}

.vz-colour-row .vz-field { flex: 1 1 auto; margin: 0; }

/* ⚠ A NATIVE `<input type="color">`, WHICH IS E24'S DECISION APPLIED TO A SECOND
   CONTROL. The swatch it draws is the browser's and cannot be restyled beyond its box;
   the picker it opens is the operating system's. Both accepted for E24's reason — a
   hand-built colour wheel is a large, hard-to-reach widget, and this one is optional
   because the hex field beside it does the same job. */
.vz-colour-input {
  width: var(--vz-tap-min);
  height: var(--vz-tap-min);
  flex: none;
  padding: 2px;
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  cursor: pointer;
}

.vz-colour-input:focus-visible { outline: 2px solid var(--vz-primary); outline-offset: 2px; }

/* ── THE OPERATOR'S LOGO, AS IT WILL BE PRINTED (`§8a` row 7, `§7` item 25) ──
   ⚠ A BOX AROUND IT, BECAUSE A LOGO ON WHITE HAS NO EDGE. Most marks are transparent
   PNGs on a white page, so without a border a reader cannot tell where the image stops —
   and "is the whitespace mine or yours" is the one question a letterhead preview exists
   to answer.

   ⚠ AND A HEIGHT CEILING RATHER THAN A WIDTH. `BrandLogoWriter` stores at up to 480px on
   the LONGEST edge, so a tall mark and a wide one arrive at the same longest edge and
   only a height cap makes the panel a predictable size for both. `max-width: 100%` is
   `module.css`'s standing rule for an image in a flow, and it is what keeps a wide mark
   inside a 320px panel. */
.vz-logo-preview {
  margin: 0 0 var(--vz-space-3);
  padding: var(--vz-space-3);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
}

.vz-logo-preview img {
  display: block;
  max-width: 100%;
  max-height: 5rem;
}

/* --- The workspace quick start (F30) -------------------------------------- */

/* 🔴 THE EIGHT `.vz-home-*` RULES THAT WERE HERE ARE GONE WITH THE MARKUP THEY PAINTED.
   E4b wrote them, moving eleven inline `style` attributes off the S9 landing, and F30
   deletes that landing. ⚠ THEY WENT IN THE SAME COMMIT AS THE VIEW: a rule painting
   nothing is residue that makes a class look available, and `§E25` found three invented
   class names that were already `.vz-badge`, `.vz-kv` and `.vz-page-actions`.

   ⚠ NOTHING BELOW INTRODUCES A COLOUR OR A SPACING. Every value is a `--vz-*` token,
   every state carries a second signal beside its colour (WCAG 1.4.1), and nothing here
   needs JavaScript: the bar is a `<progress>`, the checklist is an `<ol>`, and every
   control is a link to a real URL. */

.vz-quickstart { max-width: 52rem; }

.vz-quickstart-head { margin-bottom: var(--vz-space-6); }

.vz-quickstart-title {
  margin: 0 0 var(--vz-space-2);
  font-size: var(--vz-fs-h1);
  line-height: var(--vz-lh-tight);
  color: var(--vz-primary-strong);
}

.vz-quickstart-lede {
  margin: 0 0 var(--vz-space-4);
  color: var(--vz-subtle);
  font-size: var(--vz-fs-sm);
}

/* 🐛 ⚠ THE BAR SITS IN THE REQUIRED GROUP'S HEADING ROW, AND IT USED TO HAVE A `.vz-
   quickstart-meter` OF ITS OWN IN THE PAGE HEADER. That printed "1 of 5 done" twice,
   thirty pixels apart, with every assertion green and the 320→1440 sweep at 0px — found
   by looking at the screen. The wrapper went with the duplicate.

   ⚠ `appearance: none` AND BOTH VENDOR PSEUDO-ELEMENTS, BECAUSE A `<progress>` IS DRAWN
   BY THE PLATFORM OTHERWISE — grey-on-grey in one browser and a system green in another,
   neither of them this product's colour. The four rules here are ONE control painted
   three ways: the element itself is Firefox's track, `::-webkit-progress-bar` is the
   track everywhere else, and the two value pseudo-elements are the fill. They cannot be
   combined into one selector list: a browser drops the whole rule when it meets a
   pseudo-element it does not know. */
.vz-quickstart-bar {
  -webkit-appearance: none;
  appearance: none;
  /* ⚠ `min-width: 0` IS THE GROWABLE ITEM'S ESCAPE (`§E25` finding 10). A flex item's
     floor is its content size, and a `<progress>` carries its own fallback text — so
     without this the bar would refuse to go below the width of "3 of 5 done".
     ⚠ AND THE CAP IS SO IT STAYS A GAUGE. Uncapped it grew to the whole 52rem column,
     which turns a five-step meter into a rule across the page. */
  flex: 1 1 6rem;
  min-width: 0;
  max-width: 14rem;
  height: 0.5rem;
  border: 0;
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface-2);
}

.vz-quickstart-bar::-webkit-progress-bar {
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface-2);
}

.vz-quickstart-bar::-webkit-progress-value {
  border-radius: var(--vz-radius-pill);
  background: var(--vz-primary-button);
}

.vz-quickstart-bar::-moz-progress-bar {
  border-radius: var(--vz-radius-pill);
  background: var(--vz-primary-button);
}

.vz-quickstart-group { margin-bottom: var(--vz-space-6); }

/* ⚠ `center` RATHER THAN `baseline`, because a `<progress>` has no text in it to sit a
   baseline on — aligned that way it drops to the bottom of the heading's line box. */
.vz-quickstart-group-head {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-3);
}

.vz-quickstart-group-title {
  margin: 0;
  font-size: var(--vz-fs-md);
  color: var(--vz-ink);
}

/* ⚠ A CHIP AND NOT A SECOND BAR. Two progress bars on one screen invite the reader to
   add them up, and the number they would arrive at is exactly the combined percentage
   F30 rules out by name: an operator at 100% required and 0% recommended is done.

   🔴🐛 `margin-left: auto` — AND IT IS AN ALIGNMENT FIX FOUND BY DRIVING THE SCREEN
   (F30's drive, 2026-09-24). The two chips say the same kind of thing about the two
   groups and they are read as a column, so they have to BE one. Left-packed they are
   not: the chip trails whatever precedes it, and what precedes it differs in both
   groups — *Required* has a bar and an eight-character word, *Recommended* has no bar
   and an eleven-character one. **Measured at x=576 and x=384, 192px apart, one
   directly under the other.**

   ⚠ NOTHING COULD HAVE MEASURED THAT AS WRONG. Both chips were the right size, inside
   their head, with no overflow at any width — `.vz-result-bar`'s own title-left /
   meta-right shape is the pattern this now matches. */
.vz-quickstart-group-count {
  margin-left: auto;
  padding: var(--vz-space-1) var(--vz-space-2);
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface-2);
  color: var(--vz-muted);
  font-size: var(--vz-fs-xs);
  font-variant-numeric: tabular-nums lining-nums;
}

.vz-quickstart-group-note {
  margin: var(--vz-space-2) 0 var(--vz-space-3);
  color: var(--vz-subtle);
  font-size: var(--vz-fs-sm);
}

/* ⚠ `overflow: hidden` IS WHAT MAKES THE CURRENT STEP'S TINT RESPECT THE CORNER. A
   tinted first or last child otherwise paints square over the container's radius, which
   reads as a rendering fault rather than as a highlight. */
.vz-quickstart-list {
  margin: 0;
  padding: 0;
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  list-style: none;
  overflow: hidden;
}

.vz-quickstart-step {
  display: flex;
  align-items: flex-start;
  gap: var(--vz-space-3);
  padding: var(--vz-space-4);
}

/* ⚠ THE ADJACENT-SIBLING SELECTOR RATHER THAN A BORDER PLUS A `:first-child` THAT TAKES
   IT BACK. Two top-level rules setting the same property on one class is the shape
   `testNoClassIsSilentlyRedefinedByASecondTopLevelRule` exists for; this way the border
   is declared once and there is nothing to overrule. */
.vz-quickstart-step + .vz-quickstart-step { border-top: 1px solid var(--vz-border); }

/* 🔴 THE NEXT UNDONE STEP. Tint, weight and a filled mark — three signals, so the
   highlight survives a reader who cannot distinguish the colour. */
.vz-quickstart-step.is-current { background: var(--vz-primary-tint); }

.vz-quickstart-step.is-current .vz-quickstart-step-title { font-weight: var(--vz-fw-bold); }

.vz-quickstart-mark {
  display: inline-flex;
  flex: 0 0 auto;
  align-items: center;
  justify-content: center;
  width: 2rem;
  height: 2rem;
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface-2);
  color: var(--vz-muted);
}

/* ⚠ THE BUTTON FILL AND ITS CONTRAST TOKEN, which is what `.vz-step.is-current
   .vz-step-no` does for the wizard's current step. `--vz-primary-strong` is the token
   for text ON a surface and would be the wrong one to put a glyph on top of. */
.vz-quickstart-step.is-current .vz-quickstart-mark {
  background: var(--vz-primary-button);
  color: var(--vz-primary-contrast);
}

.vz-quickstart-step.is-done .vz-quickstart-mark {
  background: var(--vz-success-tint);
  color: var(--vz-success-text);
}

/* A finished step is quieter than an outstanding one and is never struck through:
   the work stands, and crossing it out says it was cancelled. */
.vz-quickstart-step.is-done .vz-quickstart-step-title { color: var(--vz-muted); }

.vz-quickstart-body {
  flex: 1 1 auto;
  /* The growable item's escape again — a long step title must wrap rather than widen
     the row past the container. */
  min-width: 0;
}

.vz-quickstart-step-title {
  margin: 0;
  color: var(--vz-ink);
  font-size: var(--vz-fs-base);
  font-weight: var(--vz-fw-medium);
}

.vz-quickstart-step-line {
  margin: var(--vz-space-1) 0 0;
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

.vz-quickstart-acts {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2);
  margin-top: var(--vz-space-3);
}

/* --- Motion (E13) --------------------------------------------------------- */

/* ⚠ THE OWNER ASKED FOR THIS IN THESE WORDS: *"Actions and interactions should feel
   natural and effortless, with subtle transitions and hover effects."* Every hover
   state below already existed and every one of them SNAPPED — the product had exactly
   one transition in it, on the focus ring of a form control.

   ⚠ COLOUR ONLY, AND ON PURPOSE. Nothing here transitions a size, a position or a
   shadow that would move the box: a dense table would shimmer as the pointer crossed
   it, and a row that grows on hover is a row that is hard to click on a phone. What
   changes is what the element is painted with, which is exactly what a hover state
   means in this design system.

   ⚠ AND `outline` IS DELIBERATELY EXCLUDED. `:focus-visible` is the one indicator a
   keyboard user depends on and it has to arrive instantly; fading it in is a delay on
   the exact affordance that must not have one. */
/* ⚠ `.vz-th-sort` IS THE ANCHOR, NOT A WRAPPER AROUND ONE. The views build it as
   `class="vz-th-sort ..."` on the `<a>` itself, so a `.vz-th-sort a` selector here would
   match nothing at all and the sortable headers alone would keep snapping — a whole
   family of controls silently left out, with nothing to see and nothing to fail. */
/* 🔴 **A BUTTON THAT HAS TO READ AS A LINK, AND UNTIL 2026-10-13 IT READ AS NEITHER.**
   Owner: *"in the lock screen signout link desing not looks good"*. The class carried only the
   `transition` below and nothing that paints, so it rendered as the browser's native grey
   button in the middle of an auth card.

   ⚠ **IT IS A `<button>` BECAUSE SIGNING OUT IS A POST**, not because anybody wanted a button:
   a GET that ends a session is re-fired by a prefetch or a link preview. ▶ So the element is
   correct and only its appearance was missing — which is why this resets the native chrome
   rather than inventing a new control.

   🔴 **AND IT MATTERS MORE HERE THAN IT LOOKS.** `auth/locked.php` says it: somebody who has
   walked away from a shared branch terminal must always be able to end the session rather than
   being forced to remember a password first, *"otherwise the safest action is the one the
   screen makes hardest"*. A control that looks like a misplaced form widget is that, by
   accident.

   ⚠ **UNDERLINED AT REST, NOT ONLY ON HOVER.** It sits alone under a card with no surrounding
   link text to borrow its affordance from, and hover is not available to a touch reader at all.
   ⚠ **AND IT KEEPS A REAL TAP TARGET** — this is a telephone screen as often as a desktop one,
   and `--vz-tap-min` is the floor the rest of the product uses. */
.vz-linkbtn,
.vz-footer a,
.vz-th-sort {
  transition: background-color var(--vz-transition),
              border-color var(--vz-transition),
              color var(--vz-transition);
}

/* ⚠ NOT OPTIONAL, AND NOT A COURTESY. `prefers-reduced-motion` is a real setting a real
   user turns on because movement makes them ill — vestibular disorders, migraine — and a
   product that adds transitions without honouring it has made itself worse for exactly
   the people who took the trouble to ask. The duration is near-zero rather than `none`
   so that any `transitionend` handler still fires; nothing in this product listens for
   one today, and this way nothing breaks the day something does. */
@media (prefers-reduced-motion: reduce) {
  *,
  *::before,
  *::after {
    transition-duration: 0.01ms !important;
    animation-duration: 0.01ms !important;
    animation-iteration-count: 1 !important;
    scroll-behavior: auto !important;
  }
}

/* --- Inline icons (E14) --------------------------------------------------- */

/* ⚠ `currentColor` IS THE WHOLE MECHANISM. An icon takes the colour of the text it sits
   beside, so the alert icons match their own ink and the button icon matches its label
   without one rule per variant. `1em` sizing means an icon scales with its context
   rather than needing a size class per placement. `partials/icon.php` explains why these
   are paths rather than the icon FONT the finance portal uses. */
.vz-icon {
  width: 1.15em;
  height: 1.15em;
  flex: none;
  vertical-align: -0.2em;
}

/* 🔴 THE ICON SPRITE'S DEFINITIONS BLOCK (row 13, 2026-09-28).

   `App\Support\IconSprite` writes one `<symbol>` per DISTINCT glyph just before `</body>`,
   and every icon on the page became a `<use>` reference to it. Measured on
   `/app/audit?per_page=250` beforehand: 306 icons, 37 distinct, 98,448 bytes of inline
   `<svg>` — 35.5% of the response, on a product whose compression is HELD (`F22`), so raw
   bytes are wire bytes.

   ⚠ `display: none` RATHER THAN THE `hidden` ATTRIBUTE, because the attribute is a
   presentation hint an author stylesheet can lose; this is the stylesheet. A `<symbol>` is
   never painted anyway — this hides the `<svg>` element that CONTAINS them, which otherwise
   occupies a line box at its default 300×150. */
.vz-sprite { display: none; }

/* 🔴 **THE MARK THAT A LINK LEAVES FOR A NEW TAB** (owner, 2026-10-13). The operator
   dashboard's three links all open in one, and this is the only thing on the page that says
   so to a reader who can see — the glyph is `aria-hidden` like every icon here, so
   `app.dashboard.new_tab` says it in words beside each one.

   ⚠ **MUTED, NOT `currentColor`, WHICH IS THIS FILE'S ONE DELIBERATE EXCEPTION TO ITS OWN
   MECHANISM.** Every other glyph takes the colour of the text it sits beside, and that is
   exactly wrong here: beside a red overdue count the arrow would read as part of the FIGURE,
   and beside a widget title as part of the NAME. ▶ It is not about the thing it sits next to,
   it is about what pressing it does, so it is deliberately the quietest ink on the row. */
.vz-icon--open { color: var(--vz-muted); }

/* The reveal button ships BOTH icons and shows one — so the swap needs no markup
   built in JavaScript, and the icon is correct before the script has even run. */
.vz-control-addon .vz-icon--hide,
.vz-control-addon.is-revealed .vz-icon--show { display: none; }
.vz-control-addon.is-revealed .vz-icon--hide { display: block; }


/* ===========================================================================
   Staff: the register (P1)
   ---------------------------------------------------------------------------
   ⚠ THE MODIFIER IS LOAD-BEARING, NOT DECORATION — the rule every table in this
   file since B5 has had to be told. At phone width the generic rules turn a
   `.vz-table` row into a CSS grid whose areas name the MEMBER columns
   (`.vz-c-name`, `.vz-c-code`, `.vz-c-phone`, `.vz-c-kyc`, `.vz-c-joined`). A
   staff row carries `.vz-c-role` and `.vz-c-branches`, which match no area at
   all, so without this block they are auto-placed into implicit rows in an
   order nobody chose — **and the horizontal-overflow measurement still reads
   ZERO, because they fit.** That is exactly how B5 shipped two broken tables,
   so this block goes in with the markup rather than after somebody notices.

   ⚠ AND `actions` IS A NAMED AREA HERE FOR THE SAME REASON `UI_CONVENTIONS §2a`
   gives: *"A NEW ACTIONS COLUMN NEEDS A NAMED `actions` AREA IN THAT TABLE'S
   CARD GRID … check the computed `grid-area` of every cell at 375, not the
   overflow number."*
   =========================================================================== */

/* 🐛🔴 THE RULE HERE USED TO CONTRADICT THE SENTENCE ABOVE IT, AND THE SENTENCE WAS RIGHT.
   The comment reads — and still reads — that *"a role and a branch list are ordinary prose and
   must be free to wrap inside their cell rather than force the table wider"*, and the next
   line was `.vz-c-role { white-space: nowrap; }`. **It did exactly what its own reason says it
   must not.** Measured at 1025 on `/app/users`: letting this one column wrap takes the table
   **43px narrower**, which is half of what that register needed.
   ▶ Found by item 63's row, by measuring which column each table's overflow was actually in
   and then reading the rule that produced it. ⚠ **The login id is the half that was correct**:
   it is dictated down a phone line, it inherits `.vz-c-code`'s `nowrap`, and it keeps it. */
.vz-c-branches { color: var(--vz-muted); font-size: var(--vz-fs-sm); }

@media (max-width: 64rem) {
  /* The card leads with WHO and WHETHER THEY ARE ACTIVE — the two things
     somebody scans a staff list for — then the credential and the role, then
     where they may work. `added` sits with the actions row because the date a
     colleague joined is the last thing anybody looks for on this screen. */
  .vz-table--users tr {
    grid-template-areas:
      'name     status'
      'code     role'
      'branches branches'
      'joined   joined'
      'actions  actions';
  }

  /* ⚠ `width: auto` UNDOES THE SHRINK-TO-FIT, exactly as the schedule's cells
     do: this is a grid area now, not a table cell, and 1% of a card is 3px.
     `white-space: normal` is the other half — a branch list of three names
     cannot break inside a `nowrap` cell and would push the page sideways at
     320px, which is the 99px defect `§2a` records against the members list. */
  .vz-table--users .vz-c-role,
  .vz-table--users .vz-c-branches {
    width: auto;
    white-space: normal;
  }

  .vz-table--users .vz-c-role     { grid-area: role; justify-self: end; text-align: right; }
  .vz-table--users .vz-c-branches { grid-area: branches; }
}


/* --- A printed credential (P1) --------------------------------------------- */

/* 🐛 ⚠ THIS EXISTS BECAUSE `.vz-facts` MEASURED WRONG, AND THE MEASUREMENT IS THE WHOLE
   ARGUMENT. The create outcome first rendered the login id and the temporary password as
   a `<dl class="vz-facts">`, which is `minmax(8rem, 14rem) 1fr`: at 320px the label column
   takes 128px and the value takes what is left, and a 16-character password is **one
   unbreakable token with a min-content width of 127px**. Measured: **25px of document
   overflow at 320px**, and clean at every wider width — the exact shape of defect a
   three-fixed-widths check misses and a full 320→1440 sweep does not.

   ⚠ THE OTHER FIX WOULD HAVE BEEN TO COLLAPSE `.vz-facts` TO ONE COLUMN ON A PHONE, and it
   was rejected: that grid draws ten panels on the member profile and several more
   elsewhere, and restyling all of them to fix one screen is a design change this row was
   not asked to make.

   ⚠ AND THE SMALLER FIX IS ALSO THE BETTER ONE. A credential is not a fact in a list — it
   is the thing somebody is about to read down a phone line — so the label sits ABOVE it
   and the value gets the whole width at a size that can be dictated without misreading a
   1 for an l. `overflow-wrap: anywhere` is what makes it structurally unable to overflow
   at any width rather than merely wide enough today. */
.vz-credential {
  display: flex;
  flex-direction: column;
  gap: var(--vz-space-1);
  margin: var(--vz-space-3) 0;
}

.vz-credential-label {
  color: var(--vz-subtle);
  font-size: var(--vz-fs-xs);
  text-transform: uppercase;
  letter-spacing: 0.08em;
}

.vz-credential-value {
  font-family: var(--vz-font-mono);
  font-size: var(--vz-fs-lg);
  letter-spacing: 0.08em;
  color: var(--vz-ink);
  /* The one declaration that makes the overflow impossible rather than unlikely. */
  overflow-wrap: anywhere;
  /* A whole credential in one click, because the alternative is retyping it by eye. */
  user-select: all;
}


/* ===========================================================================
   Roles: the register, and the permission matrix (P2)
   ---------------------------------------------------------------------------
   ⚠ THE MATRIX IS THE LARGEST CONTROL SURFACE IN THE PRODUCT — sixteen modules
   × five verbs, plus seven abilities — and it is the only screen where losing
   your place costs somebody a permission. Everything below is in service of one
   property: **a cell must be unambiguous about which module and which verb it
   belongs to, at every width**, because the failure mode here is not an ugly
   page, it is an operator granting `payouts.delete` while looking at the
   `payments` row.
   =========================================================================== */

/* The register. Two columns shrink to their content: a "Supplied"/"Yours" badge
   and a holder count are both narrow and fixed, and `.vz-table`'s `width: 100%`
   would otherwise hand them the slack and float them away from their heading —
   `.vz-c-actions`' own reason (E6b), applied to the two columns beside it. */
.vz-c-kind    { width: 1%; white-space: nowrap; }
.vz-c-holders {
  /* ⚠ F38 took `text-align: right` out — see the base table section. */
  width: 1%;
  white-space: nowrap;
  font-variant-numeric: tabular-nums lining-nums;
}

/* ⚠ THE MATRIX'S FIRST COLUMN IS A `<th scope="row">`, NOT A `<td>`, and this is
   what stops it looking like one. `.vz-table th` is styled for the HEAD row;
   a row header inherits that and would render every module name in the header's
   weight and colour, which reads as sixteen sub-headings rather than sixteen
   rows. */
.vz-table--matrix .vz-c-module {
  width: 40%;
  text-align: left;
  font-weight: var(--vz-fw-regular);
  color: var(--vz-ink);
  white-space: normal;
}

/* ⚠ CENTRED, AND `width: 1%` SO THE FIVE VERB COLUMNS ARE EQUAL AND NARROW. A
   checkbox left-aligned in a wide column sits nowhere near its heading, and on a
   grid this dense the only thing keeping a tick attached to its verb is that the
   column is no wider than the heading above it. */
.vz-table--matrix .vz-c-verb {
  width: 1%;
  text-align: center;
  white-space: nowrap;
}

/* The whole cell is the target, not just the 20px box — `--vz-tap-min` in both
   directions, which `.vz-action--row`'s comment argues for at length. */
.vz-matrix-cell {
  display: flex;
  align-items: center;
  justify-content: center;
  gap: var(--vz-space-2);
  min-height: var(--vz-tap-min);
  cursor: pointer;
}

/* 🔴 ⚠ CLIPPED, NOT `display: none`, AND THE TWO HALVES OF THE MATRIX NEED
   OPPOSITE THINGS FROM IT.

   On an EDITABLE cell this span duplicates the input's `aria-label`, so removing
   it from the tree would be right — `.vz-action-label`'s rule. On a READ-ONLY
   cell it is the **only text in the cell**: the glyph is `aria-hidden`, so
   `display: none` here would leave a cell that announces as blank, which is the
   absent-accessible-name defect rather than the duplicated one.

   Clipping serves both. The read-only cell keeps its sentence in the tree at
   every width; the editable cell's duplicate is announced beside its label,
   which is WCAG 2.5.3's *"label in name"* satisfied rather than violated —
   the visible word is a substring of the accessible name, in the same order.
   ⚠ `white-space: nowrap` COMES WITH THE TECHNIQUE and must stay: without it a
   clipped 1px box wraps its text to one character per line and some screen
   readers announce it letter by letter. */
.vz-matrix-verb {
  position: absolute;
  width: 1px;
  height: 1px;
  margin: -1px;
  padding: 0;
  overflow: hidden;
  clip: rect(0 0 0 0);
  clip-path: inset(50%);
  white-space: nowrap;
  border: 0;
}

/* The read-only cell: a glyph, with the sentence beside it in `.vz-sr-only`. */
.vz-matrix-state {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  min-height: var(--vz-tap-min);
  color: var(--vz-muted);
}

/* ⚠ `--vz-primary-strong` AND NOT `--vz-primary`. This glyph is the only thing
   distinguishing a granted cell from an ungranted one on the read-only screen,
   so it is carrying meaning rather than decorating — and `--vz-primary` is held
   to a NON-TEXT floor (`theme.css`), which is the exact distinction `§E12`
   found the product's own name failing. */
.vz-matrix-state.is-on { color: var(--vz-primary-strong); }

/* One ability per line on the read-only screen, reading as a sentence with its
   mark in front of it. */
/* 🔴 `F50` — **THIS ROW COULD NOT BREAK, AND IT PUSHED THE PAGE 46px SIDEWAYS AT 320.**
   Measured on `/app/roles/{id}`: `<span class="vz-badge">Asks for a password</span>` is a
   direct flex child here, 93px wide, and it started at **x = 273 in a container that ends at
   268** — outside its own parent, on a page whose `scrollWidth` was therefore 366 against a
   320px viewport.

   ⚠ **THE SIBLING RENDERING PROVES THE CAUSE RATHER THAN ARGUING IT.** `_matrix.php` draws
   an ability two ways: the EDITABLE one nests the name and the badge together inside a
   `<span>` in the `<label>`, so they are inline content that wraps by itself, and
   `/app/roles/{id}/edit` measured **0px at every width**. The READ-ONLY one makes the badge a
   flex item of this row, and a flex line that cannot wrap has nowhere to put it. **Two
   renderings of one thing, and only the one without an inline wrapper overflows.**

   🐛 ▶ **AND `flex-wrap: wrap` ALONE WAS THE WRONG FIX, WHICH ONLY LOOKING SHOWED.** It
   cleared the 46px — and then pushed a long ability (*"Void or bounce a payment somebody
   holds a receipt for"*) onto its own line, **leaving the state icon stranded above it with a
   blank row beside it**. That is `F14`'s orphaned value arriving as the cure. ▶ So the view
   now wraps the name, the screen-reader reading and the badge in ONE `<span>`, exactly as the
   editable `<label>` has always done: inline content that wraps in place beside an icon that
   stays put. `min-width: 0` is what lets that span be narrower than its longest word's line
   box, and it is the declaration doing the work. */
.vz-matrix-ability {
  display: flex;
  align-items: baseline;
  gap: var(--vz-space-2);
  margin: 0 0 var(--vz-space-2);
  color: var(--vz-muted);
}

.vz-matrix-ability-text { min-width: 0; }

.vz-matrix-ability.is-on { color: var(--vz-ink); }

/* ⚠ THE ABILITIES ARE A LIST OF CHECKBOXES, NOT A GRID. There are seven of them
   and each carries a whole sentence — "Void or bounce a payment somebody holds a
   receipt for" — so a column layout would either truncate them or produce seven
   columns of one word. `.vz-check` already owns the control itself; this is only
   the spacing between them, and it carries the room BELOW itself (`§2b`). */
.vz-check-list {
  display: flex;
  flex-direction: column;
  gap: var(--vz-space-3);
  margin: 0 0 var(--vz-space-5);
}

@media (max-width: 64rem) {
  /* The register's card. A role is read as "what is it called, what is it for",
     and the holder count decides whether the actions below are offered at all — so it
     shares the name's row rather than sitting under it.

     🐛🔴 F16 — THIS TEMPLATE OUTLIVED THE MARKUP IT WAS WRITTEN FOR. It declared FIVE
     areas — `name`, `note`, `kind`, `holders`, `actions` — across four rows, and the view
     renders THREE cells: `.vz-c-name`, `.vz-c-holders`, `.vz-c-actions`. `.vz-c-note` and
     `.vz-c-kind` no longer exist; the view's own comment records why —
     *"THE DESCRIPTION AND THE KIND FOLD INTO THE NAME CELL, which is `§Q9`'s own fix on
     `/admin/users`"*. The stylesheet was not told.

     ▶ THE VISIBLE COST: `holders` was placed in the right half of `"kind holders"`, a row
     whose left half nothing occupied — so *"Nobody"* rendered at x = 758 against empty
     space, floating between the description and the buttons with no column to belong to.
     ⚠ An unused `grid-template-areas` name is SILENT: nothing warns, nothing overflows, and
     the sweep measured zero. It only reads as wrong. */
  .vz-table--roles tr {
    grid-template-areas:
      "name    holders"
      "actions actions";
  }

  .vz-table--roles .vz-c-name    { grid-area: name; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--roles .vz-c-holders { grid-area: holders; width: auto; justify-self: end; }
  .vz-table--roles .vz-c-actions { grid-area: actions; text-align: left; }

  /* 🔴 ⚠ THE MATRIX DOES NOT BECOME A CARD GRID, AND THAT IS THE ONE DELIBERATE
     EXCEPTION IN THIS FILE. Every other table names an area per cell because the
     alternative is auto-placement in an order nobody chose. Here the columns are
     KNOWN and there are six of them, five of which are a single checkbox — and a
     `"create read" "update delete"` card would put two ticks side by side with
     their words at different lengths, which is precisely how somebody grants
     `delete` while reading `update`.

     So the row becomes a BLOCK: the module name on its own line, then the five
     verbs stacked, each one a checkbox with its word beside it. It is taller and
     it is unambiguous, and on the one screen in the product where a misread
     costs a permission that is the correct trade. `.vz-table--report` already
     opts out of the card layout for its own reason, so this is the second and it
     is not a new idea. */
  .vz-table--matrix tr {
    display: block;
    margin: 0 0 var(--vz-space-3);
    padding: var(--vz-space-3);
    border: 1px solid var(--vz-border);
    border-radius: var(--vz-radius-md);
    background: var(--vz-surface);
  }

  .vz-table--matrix .vz-c-module {
    display: block;
    width: auto;
    padding: 0 0 var(--vz-space-2);
    border: 0;
    font-size: var(--vz-fs-md);
    font-weight: var(--vz-fw-bold);
  }

  /* ⚠ FLEX ON A `<td>` HERE IS NOT `§2a-i`'s DEFECT, AND THE DIFFERENCE IS THE
     WIDTH. That rule is about a cell inside a real table, where taking it out of
     the table box model breaks the row's shared bottom edge and collapses the
     column's min-content. Below 44rem `.vz-table td` is ALREADY `display: block`
     — there is no table box model left to leave, which is why
     `testNoTableCellIsTakenOutOfTheTableBoxModel` cuts this media block out by
     brace matching rather than exempting it. The glyph and its words are one
     line, so they are a flex row. */
  .vz-table--matrix .vz-c-verb {
    display: flex;
    align-items: center;
    gap: var(--vz-space-2);
    width: auto;
    min-height: var(--vz-tap-min);
    padding: 0;
    border: 0;
    text-align: left;
  }

  /* ⚠ `flex-start` RATHER THAN `center`: at this width the label is a real word
     beside the box, so the pair reads left to right like every other checkbox on
     every other form in the product. */
  .vz-table--matrix .vz-matrix-cell { justify-content: flex-start; }

  /* ⚠ THE WORD COMES BACK, AND ON THE READ-ONLY MATRIX IT IS THE WHOLE CELL. The
     column headings are gone at this width, so a bare tick says nothing to
     anybody — sighted or not. Every property the clipping technique set has to be
     undone by name; a single `position: static` would leave a 1px box. */
  .vz-matrix-verb {
    position: static;
    width: auto;
    height: auto;
    margin: 0;
    overflow: visible;
    clip: auto;
    clip-path: none;
    white-space: normal;
  }

  /* The glyph is no longer a column of its own; the cell above supplies the tap
     height, so the mark stops carrying one and simply sits beside its words. */
  .vz-table--matrix .vz-matrix-state { min-height: 0; }
}


/* ===========================================================================
   Branches: the register (P3)
   ---------------------------------------------------------------------------
   ⚠ THE ONE REGISTER IN THE PRODUCT THAT SHOWS CLOSED ROWS BY DEFAULT, so the
   status column is doing more work here than on any other list: it is the
   difference between a branch somebody can work in and one that is only on the
   record. It gets a badge and nothing else — a greyed-out ROW would say the same
   thing in colour alone, and a closed branch is not disabled content.
   =========================================================================== */

/* A state is an ISO code — `IN-TN` — or the words "Not set". Neither wraps, and
   both are narrow, so the column shrinks to its content the way `.vz-c-kind`
   does on the roles register. */
.vz-c-state { width: 1%; white-space: nowrap; }

@media (max-width: 64rem) {
  /* A branch is read as "which one, and can I work there" — so the card leads
     with the name and puts the code beside the status, which are the two facts
     that answer the second question. */
  .vz-table--branches tr {
    grid-template-areas:
      "name    name"
      "code    status"
      "state   opened"
      "actions actions";
  }

  .vz-table--branches .vz-c-name    { grid-area: name; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--branches .vz-c-code    { grid-area: code; }
  .vz-table--branches .vz-c-status  { grid-area: status; justify-self: end; text-align: right; }
  .vz-table--branches .vz-c-state   { grid-area: state; width: auto; }
  .vz-table--branches .vz-c-joined  { grid-area: opened; justify-self: end; text-align: right; }
  .vz-table--branches .vz-c-actions { grid-area: actions; text-align: left; }
}


/* ===========================================================================
   The audit trail (P4)
   ---------------------------------------------------------------------------
   ⚠ THE ONLY UNBOUNDED REGISTER IN PHASE P, and the only one that carries
   `§B1`'s full list grammar. Roles and branches are read in one screenful;
   this one gains a row every time anybody does anything, so every column here
   is sized for scanning hundreds rather than reading eight.
   =========================================================================== */

/* ⚠ THE ACTION IS THE COLUMN SOMEBODY READS DOWN, so it takes the slack and
   wraps. Everything else on this row is short and fixed. */
.vz-c-action { white-space: normal; color: var(--vz-ink); }

/* A branch name, or "The whole organization" for an org-level act. It shrinks
   to its content — `.vz-c-kind`'s reason on the roles register: `.vz-table` is
   `width: 100%` and would otherwise hand a narrow column all the slack. */
.vz-c-branch { width: 1%; white-space: nowrap; color: var(--vz-muted); }

/* ⚠ STORED JSON, SHOWN AS IT WAS WRITTEN. A `<pre>` inherits the UA monospace
   at a size nothing else on the page uses, and `overflow-x: auto` is what stops
   one long line of new values pushing the whole page sideways — the defect
   `.vz-table-wrap` exists for, arriving in an element that is not a table.

   ⚠ AND IT IS A CONTAINING BLOCK, for `§2a`'s reason: anything with
   `overflow-x: auto` must be, or an absolutely-positioned descendant resolves
   against the initial containing block and escapes the scroll. */
.vz-audit-values {
  position: relative;
  overflow-x: auto;
  margin: 0;
  padding: var(--vz-space-3);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface-2);
  color: var(--vz-ink);
  font-family: var(--vz-font-mono);
  font-size: var(--vz-fs-sm);
  line-height: 1.5;
  /* A JSON blob has no spaces worth breaking at, so it scrolls rather than
     wrapping mid-token — a value split across two lines is a value somebody
     copies wrongly. */
  white-space: pre;
}

@media (max-width: 64rem) {
  /* A trail entry is read as one sentence: what happened, then who and when,
     then where. The action leads because it is the only variable part — every
     other cell on a card is context for it. */
  .vz-table--audit tr {
    grid-template-areas:
      "action  action"
      "who     when"
      "module  branch"
      "actions actions";
  }

  .vz-table--audit .vz-c-action  { grid-area: action; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--audit .vz-c-name    { grid-area: who; white-space: normal; }
  .vz-table--audit .vz-c-when    { grid-area: when; justify-self: end; text-align: right; }
  .vz-table--audit .vz-c-module  { grid-area: module; }
  .vz-table--audit .vz-c-branch  { grid-area: branch; width: auto; justify-self: end; text-align: right; white-space: normal; }
  .vz-table--audit .vz-c-actions { grid-area: actions; text-align: left; }
}

/* ===========================================================================
   The control plane's fault log (Q2 / N2) — the FIRST /admin register
   ---------------------------------------------------------------------------
   ⚠ THE FIRST RULES IN THIS FILE WRITTEN FOR `/admin`, AND THEY ADD NO
   VOCABULARY. Owner, 2026-09-08: the control plane reuses the operator app's
   design system and must not start a second component set. Every cell below is
   an existing `.vz-c-*` class doing the job it already does — `.vz-c-problem`
   from the onboarding row register, `.vz-c-ref`, `.vz-c-name`, `.vz-c-when`,
   `.vz-c-status`, `.vz-c-actions` — with ONE exception, argued below.

   ⚠ AND A `grid-template-areas` BLOCK IS REQUIRED RATHER THAN OPTIONAL. Below
   44rem every `.vz-table` row becomes a card laid out by named areas, and
   `§E6d` measured what an unnamed cell costs: the roster named `standing`,
   `.vz-c-standing` had no `grid-area`, the cell was auto-placed, and it pushed
   the page 32px at 320 while every other measurement read clean.
   =========================================================================== */

/* 🔴 THE ONE NEW CELL CLASS, AND THE THREE CANDIDATES IT WAS PREFERRED OVER ARE
   NAMED SO THIS DOES NOT READ AS LAZINESS.

   `occurrence_count` is how many times one distinct fault has fired. It is a
   PRIMARY triage signal — "what is breaking most" is the second question this
   screen answers and the only one this column can answer at all.

   · `.vz-c-total` is a MONEY class: `.vz-table--dues`, `--collect` and
     `--arrears` each pair it with `.vz-c-money`, and it carries the money
     colour and alignment. Reaching for it because it means "a total" is
     `OPERATING_MODE §7` item 54 exactly — the name that is in the list rather
     than the name that says what the thing is.
   · `.vz-c-no` is a 3rem ORDINAL cell in `--vz-subtle`: right for an
     installment number beside the row it labels, wrong for a figure somebody
     sorts the table by. Subtle grey would understate the loudest fault on the
     screen.
   · `.vz-c-money` is money.

   So: tabular figures, so 1 and 7 do not shift the column when the list is
   sorted by it, and `--vz-ink` because it is read rather than skimmed past. */
.vz-c-hits {
  /* ⚠ F38 took `text-align: right` out — see the base table section. */
  width: 1%;
  white-space: nowrap;
  font-variant-numeric: tabular-nums lining-nums;
  color: var(--vz-ink);
}

@media (max-width: 64rem) {
  /* The card leads with the PROBLEM, because that is what a person scans for at
     23:00 with a phone in one hand — and the reference and the route ride inside
     that cell rather than beside it (see the view). The count and the moment share
     a line, since "how many" and "how recently" are read together. */
  .vz-table--errors tr {
    grid-template-areas:
      'problem problem'
      'org     status'
      'hits    when'
      'actions actions';
  }

  .vz-table--errors .vz-c-problem { grid-area: problem; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--errors .vz-c-name    { grid-area: org; white-space: normal; }
  .vz-table--errors .vz-c-status  { grid-area: status; justify-self: end; text-align: right; }
  .vz-table--errors .vz-c-hits    { grid-area: hits; width: auto; text-align: left; }
  .vz-table--errors .vz-c-when    { grid-area: when; justify-self: end; text-align: right; }
  .vz-table--errors .vz-c-actions { grid-area: actions; text-align: left; }
}

/* 🐛 🔴 THE DATE WRAPS IN THIS TABLE, AND THAT IS A MEASURED FIX RATHER THAN A
   PREFERENCE. `.vz-c-when`'s base rule is `white-space: nowrap`, which is right on
   nine registers where the date sits beside short cells. Here it was the single
   widest column — 162px of an unbreakable "8 Sep 2026, 11:46 am" — and it was the
   only relaxation that moved the table's min-content at all (737 → 683; wrapping
   the status and breaking the reference and the message each moved it zero).

   ⚠ IT WAS NOT ENOUGH ON ITS OWN, which is the finding: 683 still overflowed a
   553px wrapper, so the REFERENCE COLUMN was folded into the problem cell in the
   same change. Seven columns was the defect and seven widths were the symptom. */
.vz-table--errors .vz-c-when {
  white-space: normal;
}

/* ⚠ THE MESSAGE IS THE COLUMN THAT TAKES THE SLACK, which is `.vz-table--messaging`'s
   finding one register along and the opposite of every other list in this product.
   An exception message is a sentence — "SQLSTATE[42S02]: Base table or view not
   found: 1146 Table 'vazhai_org_1002.member_scores' doesn't exist" — and it is the
   whole reason somebody opened the screen, while every other cell here is short and
   fixed: a reference, a count, an instant, a badge.

   ⚠ `overflow-wrap: anywhere` AND NOT `word-break: break-all` — C3b's reason, which
   `.vz-c-ref` already carries: it only breaks a word when there is no other
   opportunity, and a class name like
   `App\Services\ImportCommitWriter` is one unbreakable token that would otherwise
   scroll the table sideways with the action off-screen while page overflow measured
   zero. That was the message queue's exact defect. */
.vz-table--errors .vz-c-problem {
  white-space: normal;
  overflow-wrap: anywhere;
}

/* The class is what a technical reader scans for; the message and the route are
   context beneath it. `.vz-sub-meta` already draws that relationship everywhere
   else in the product, and this is only telling it to take the whole line. */
.vz-table--errors .vz-c-problem .vz-sub-meta {
  display: block;
}

/* ===========================================================================
   The message queue (row 14)
   ---------------------------------------------------------------------------
   ⚠ THE LAST REPORT IS THE COLUMN THAT TAKES THE SLACK, and that is the
   opposite of every other register in the product. `last_error` is a sentence
   the SENDER wrote — "Template [dues_pre_3/whatsapp/en v1] is draft, not
   approved; awaiting Meta approval via Spring Edge." — and it is the whole
   reason somebody opened this screen. Every other cell here is short and
   fixed: an instant, a template code, a number, a badge.

   ⚠ AND `.vz-c-kind` IS RE-DECLARED WIDER THAN ITS BASE RULE. The shared one
   is `width: 1%; white-space: nowrap`, which is right for a one-word "kind" on
   the roles register; a template code plus the channel under it needs to be
   able to break, and at 320 the nowrap version pushed the table sideways.
   =========================================================================== */

.vz-table--messaging .vz-c-kind {
  width: auto;
  white-space: normal;
}

/* 🐛 🔴 THE RECIPIENT WRAPS, AND BOTH HALVES OF THIS RULE ARE A FIX.

   An email address is one unbreakable word — `branch.manager@coimbatore.example
   .test` measures 312px — and the first draft gave this column the shrink-to-fit
   `width: 1%; white-space: nowrap` that `.vz-c-branch` uses for a branch name.
   Two different defects came out of that, at opposite ends of the range:

   ⚠ AT 320 the phone card's `white-space: normal` found no break opportunity in
   the address and pushed the DOCUMENT 40px sideways, decaying to clean by 360.
   Found by sweeping every width rather than three — 375 and 390 both measure
   zero, which is how `§E25` finding 10's 326px defect hid.

   ⚠ AND ON A DESKTOP IT MADE THE TABLE 1107px INSIDE A 958px COLUMN — so the
   table scrolled horizontally at EVERY width from 900 to 1440 and the Retry
   button was never on screen without scrolling to find it. `.vz-table-wrap`
   clips that perfectly, so the page overflow reads ZERO and the sweep is right
   and blind: this one is only visible by measuring the TABLE against its wrap.

   `overflow-wrap: anywhere` and not `word-break: break-all` — C3b's argument,
   unchanged: it only breaks a word when there is no other opportunity, so a
   recipient NAME still wraps at its spaces. */
.vz-table--messaging .vz-c-name {
  white-space: normal;
  overflow-wrap: anywhere;
}

/* The report is the column that takes the slack, so nothing else may claim it.
   A template code is short and a channel word shorter, and letting them stretch
   would squeeze the one cell somebody came to read.

   🐛 ⚠ AND THE `min-width` PAIR IS NOT BELT AND BRACES — IT IS THE SECOND HALF
   OF THE `overflow-wrap: anywhere` ABOVE. Allowing a break anywhere makes the
   recipient column's MIN-CONTENT one character, and `width` is a suggestion the
   table algorithm may ignore when the table has to fit: at 770px it squeezed the
   address to ~1ch and rendered `branch.manager@coimbatore.example.test` as
   thirty-eight stacked letters. **Page overflow was zero the whole time** —
   `.vz-table-wrap` had room, so nothing was overflowing; the column had simply
   been crushed. Found in a screenshot at a width between the card breakpoint and
   the width where the table fits, which is the band nothing else looks at.

   A floor turns that back into the honest outcome: the column keeps a readable
   width, the table's minimum grows past the content column, and the WRAP scrolls
   — which is what a wrap is for. */
.vz-table--messaging .vz-c-kind { width: 12ch; min-width: 11ch; }
.vz-table--messaging .vz-c-name { width: 18ch; min-width: 14ch; }
.vz-table--messaging .vz-c-note { min-width: 16ch; }

/* 🐛 ⚠ AND THE BRANCH WRAPS HERE, WHICH THE SHARED RULE DOES NOT ALLOW.
   `.vz-c-branch` is `width: 1%; white-space: nowrap` on the audit trail, where
   the widest value is a branch name. This table's widest is the whole sentence
   *"The whole organization"* at 166px, and with six other columns beside it the
   table's MINIMUM came to 969px inside a 958px content column — so it scrolled
   sideways at 1440 and the Retry button sat off the edge at every desktop width.
   Wrapping costs a second line on one cell and buys the whole table fitting. */
.vz-table--messaging .vz-c-branch { white-space: normal; }

/* ⚠ THE TRY COUNT LINES UP UNDER THE BADGE rather than beside it, because it
   is a footnote to the state and not a second state. `.vz-c-moved` is already
   `display: block`; this only stops the nowrap badge and the line under it
   fighting for the same row. */
.vz-table--messaging .vz-c-status { white-space: nowrap; }

@media (max-width: 64rem) {
  /* ⚠ THE CARD LEADS WITH WHICH MESSAGE AND TO WHOM, because that is the pair
     somebody is matching against a member's complaint — then the state, then
     the sender's report underneath in full width, because it is a sentence and
     not a value. The retry sits at the bottom, on its own line, so a control
     that changes stored state is never adjacent to the thing a thumb reaches
     for while scrolling. */
  .vz-table--messaging tr {
    grid-template-areas:
      'message message'
      'to      state'
      'report  report'
      'branch  actions';
  }

  .vz-table--messaging .vz-c-kind    { grid-area: message; width: auto; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--messaging .vz-c-name    { grid-area: to; width: auto; white-space: normal; }

  /* ⚠ THE SECOND LINE WRAPS HERE AND NOWHERE ELSE. `.vz-c-moved`'s base rule is
     `nowrap`, which is right for the short "moved to 3 Sep" it was written for
     and wrong for a member's full name under their number in a 273px card. */
  .vz-table--messaging .vz-c-moved   { white-space: normal; }
  .vz-table--messaging .vz-c-status  { grid-area: state; justify-self: end; text-align: right; }
  .vz-table--messaging .vz-c-note    { grid-area: report; white-space: normal; }
  .vz-table--messaging .vz-c-branch  { grid-area: branch; width: auto; white-space: normal; }
  .vz-table--messaging .vz-c-actions { grid-area: actions; justify-self: end; text-align: right; }

  /* ⚠ THE RAISED-AT INSTANT IS DROPPED ON A PHONE, WITH AN EXPLICIT RULE AND
     NOT BY BEING LEFT UNPLACED. An unplaced cell is B5's blind spot — it still
     renders, auto-placed in a position nobody chose, and measures zero
     overflow because it fits. It is the one column a card can lose: the list is
     newest-first, so position already carries roughly when, and the exact
     instant is what the desktop table is for. */
  .vz-table--messaging .vz-c-when { display: none; }
}

/* ── Member risk score (D10) ───────────────────────────────────────────────
   The "why this score" headline: the number, the band chip, and the instant it
   was computed.

   ⚠ THE COMPUTED-AT LINE IS PART OF THE COMPONENT, NOT A DECORATION.
   `PHASE_D §D9` Q9 refuses event-driven recomputation, so this number is up to a
   day old by design — and the reason that is safe is that every D10 effect is
   advisory. A panel that showed the figure without the date would be implying a
   freshness the engine does not offer.

   ⚠ AND THE NUMBER IS NOT COLOURED BY BAND. The band chip carries the colour and
   the word; painting the digits as well would say the same thing twice and make a
   62 look like a verdict rather than a measurement. `--vz-ink` throughout. */
.vz-score {
  display: flex;
  align-items: baseline;
  flex-wrap: wrap;
  gap: var(--vz-space-3);
  margin-bottom: var(--vz-space-4);
}

.vz-score-value {
  font-size: var(--vz-fs-h1);
  font-weight: 600;
  line-height: 1;
  color: var(--vz-ink);
  font-variant-numeric: tabular-nums;
}

.vz-score-scale {
  font-size: var(--vz-fs-md);
  color: var(--vz-muted);
}

/* Its own line at every width — it is a sentence about provenance, and wrapping
   it beside the chip reads as a caption for the chip. */
.vz-score-note {
  flex-basis: 100%;
  font-size: var(--vz-fs-sm);
  color: var(--vz-muted);
}

@media (max-width: 64rem) {
  /* The factor leads: somebody reading this is answering "why", and the factor
     name is the answer. The raw input sits under it as a full-width phrase —
     "16 of 25 installments" is prose, not a figure — and the two numbers share
     the last line, which is where the arithmetic is checked. */
  .vz-table--factors tr {
    grid-template-areas:
      "factor factor"
      "input  input"
      "weight points";
  }

  .vz-table--factors .vz-c-name  { grid-area: factor; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--factors .vz-c-note  { grid-area: input; white-space: normal; }
  .vz-table--factors .vz-c-no    { grid-area: weight; }
  .vz-table--factors .vz-c-total { grid-area: points; justify-self: end; text-align: right; }
}

/* ===========================================================================
   The dashboard board (D12)
   ---------------------------------------------------------------------------
   ⚠ ORDER + WIDTH, NEVER COORDINATES. `PROJECT_PLAN §8a` cut drag-and-drop and
   resizing before the row started, and this is what that decision looks like in
   CSS: an `auto-fill` grid the widgets flow through in list order, with each
   widget spanning one, two or three of its columns. There is nothing here that
   places a widget at a row and a column, so removing one leaves no hole to
   re-pack — the list simply gets shorter.

   ⚠ TWELVE COLUMNS, BECAUSE 12 IS THE ONE NUMBER DIVISIBLE BY 2 AND BY 3. A
   half and a third have to line up against each other on the same board; on a
   6- or an 8-column grid one of the two stops being a clean fraction.

   ⚠ AND BELOW 44rem EVERY WIDGET IS FULL WIDTH, which is the whole reason
   placement is order + width. "Row 2, column 3" answers a question a phone
   never asks — the product is PWA-first and its heaviest users are collectors
   on phones.
   =========================================================================== */

.vz-board {
  display: grid;
  grid-template-columns: repeat(12, minmax(0, 1fr));
  gap: var(--vz-space-4);
  margin-bottom: var(--vz-space-5);
}

/* ⚠ `min-width: 0` IS LOAD-BEARING, NOT DEFENSIVE. A grid item's default
   `min-width: auto` is its content, so one long group name inside a widget
   would push the whole track wider than its share and send the page sideways —
   the failure `.vz-table-wrap` exists to stop one component along. */
/* ⚠ THE SHADOW IS FINANCE'S, AND IT IS THE TOKEN THIS PRODUCT DEFINED AND NEVER
   SPENT (§D12b). `--vz-shadow-sm` has sat in `theme.css` since E4 with **no rule
   in the product referencing it** — E4 took Finance's token VALUES onto Chit's own
   component names and never opened Finance's components, which is the mistake E14
   spent a whole row undoing. Finance puts it on `.vz-card` AND on `.vz-kpi`: every
   raised block on that dashboard lifts a hairline off the page.

   ⚠ `--sm`, NEVER `--md`. The larger one is this product's OVERLAY elevation — the
   dropdown, the dialog, the phone drawer, the widget panel — and a board of eight
   panels floating at overlay height reads as eight things about to be dismissed. */
.vz-widget {
  min-width: 0;
  padding: var(--vz-space-5);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-lg);
  background: var(--vz-surface);
  box-shadow: var(--vz-shadow-sm);
}

.vz-widget--full    { grid-column: span 12; }
.vz-widget--half    { grid-column: span 6; }
.vz-widget--third   { grid-column: span 4; }
/* ⚠ FOUR ACROSS, WHICH IS WHAT A KPI TILE IS. Added ahead of the tiles that use
   it so the vocabulary `BoardLayout::WIDTHS` names is complete in one place —
   a declared width with no rule renders as an unstyled block that overflows
   nothing and looks like ordinary text, which is the defect `StylesheetStaticTest`
   exists to catch. */
.vz-widget--quarter { grid-column: span 3; }

/* The heading IS the drill-through (`DEEP_DIVE D12`: entry points, not dead
   ends), so it is styled as a destination rather than as a decorated label. */
/* ⚠ `align-items: baseline` WOULD BE WRONG HERE. The glyph is a 1.15em box with
   no text baseline of its own, so baseline alignment drops it below the words;
   centring on the first line is what makes a 24-grid icon sit level with a
   heading that may wrap to two lines. */
.vz-widget-title {
  display: flex;
  align-items: center;
  gap: var(--vz-space-2);
  margin: 0 0 var(--vz-space-3);
  font-size: var(--vz-fs-md);
  font-weight: var(--vz-fw-bold);
  color: var(--vz-ink);
}

.vz-widget-title a { color: var(--vz-primary-strong); text-decoration: none; }
.vz-widget-title a:hover { text-decoration: underline; }

/* The one figure the widget exists for. Tabular so a column of them lines up,
   and `--vz-fs-display` because a dashboard is read from further away than a
   form is. */
.vz-widget-figure {
  margin: 0 0 var(--vz-space-1);
  font-size: var(--vz-fs-display);
  font-weight: var(--vz-fw-bold);
  color: var(--vz-ink);
  font-variant-numeric: tabular-nums lining-nums;
}

/* ⚠ THE SENTENCE UNDER THE FIGURE, AND IT IS NEVER OPTIONAL. A bare number
   under a heading is a number nobody can check. */
.vz-widget-note {
  margin: 0 0 var(--vz-space-3);
  font-size: var(--vz-fs-sm);
  color: var(--vz-muted);
}

/* ⚠ A LIST, NOT A TABLE, AND THAT IS DELIBERATE. Below 44rem every `.vz-table`
   in this product becomes the members card grid unless it brings its own
   modifier — so eight widgets would have meant eight modifiers for lists that
   are two or three short lines each. These are named lists, not data tables. */
.vz-widget-rows { margin: 0; padding: 0; list-style: none; }

/* 🐛🔴 F14 — THE VALUE USED TO ORPHAN ONTO ITS OWN LINE, AND THE ROW HEIGHTS RAGGED.
   This was `display: flex; flex-wrap: wrap`. When the name filled the available line the
   value had nowhere to go but the next one, and the row grew. Measured on
   `/admin/dashboard`'s Operators panel across five operators carrying IDENTICAL data shapes:

     Anand Chits            name 200px   v98 on the same line   row  61px
     Comprehensive Test Co  name 231px   v98 DROPPED            row 110px
     Heavy Usage Chits      name 231px   v98 DROPPED            row  88px
     Sri Lakshmi Chit Funds name 231px   v98 DROPPED            row  88px
     Vazhai Test Drive      name 200px   v98 on the same line   row  62px

   **61 / 110 / 88 / 88 / 62** — a list whose rhythm depended on how long a customer happened
   to be called, with `v98` stranded mid-row against nothing.

   ▶ A TWO-COLUMN GRID TAKES THE DECISION AWAY FROM THE CONTENT. `minmax(0, 1fr)` lets the
   name wrap INSIDE its own column instead of pushing its neighbour off the line, and the
   value is placed in column 2 rather than merely floated there by `margin-left: auto`. The
   meta spans both, which is what `flex-basis: 100%` was emulating.

   ⚠ IT WAS INVISIBLE UNTIL 2026-09-23. With three short-named operators every row fitted on
   one line; the two throwaway orgs seeded for the billing work are the first names long
   enough to trip it. **A defect that needed data to exist** — which is `§7` item 39's whole
   argument, and the reason a green offline sweep never found it. */
.vz-widget-rows li {
  /* 🔴 ⚠ `position: relative` EARNS ITS PLACE TWICE, AND NEITHER REASON IS DECORATION
     (2026-10-13). It is what `.vz-widget-row-link`'s stretched `::after` resolves against,
     so the whole row — including the arrow beside the figure — is the hit area.

     ⚠ AND IT IS THE FOURTH SIGHTING OF THE DEFECT `.vz-table-wrap` AND `.vz-step` BOTH
     CARRY. A `.vz-sr-only` span is `position: absolute`, and the same row now renders one
     (*"(opens in a new tab)"*); with no positioned ancestor its static position resolves
     against the INITIAL containing block, and a 1px hidden span has extended this
     product's document three times already. */
  position: relative;
  display: grid;
  grid-template-columns: minmax(0, 1fr) auto;
  align-items: baseline;
  column-gap: var(--vz-space-2);
  row-gap: var(--vz-space-1);
  padding: var(--vz-space-2) 0;
  border-top: 1px solid var(--vz-border);
}

.vz-widget-rows li:first-child { border-top: 0; }

/* Grows into the room, and CAN shrink — `min-width: 0` for the reason
   `testEveryGrowableFlexItemEitherCanShrinkOrSaysWhyNot` exists. */
/* Column 1, and it wraps inside itself rather than displacing the value (F14). The
   `min-width: 0` that `testEveryGrowableFlexItemEitherCanShrinkOrSaysWhyNot` exists for is
   now carried by `minmax(0, 1fr)` on the track, which is the grid spelling of the same
   property: a named column that is allowed to be narrower than its content. */
.vz-widget-row-name { grid-column: 1; color: var(--vz-ink); }

/* 🔴 **A ROW LINK, DRAWN THE WAY EVERY OTHER ROW LINK IN THE OPERATOR APP IS DRAWN**
   (owner, 2026-10-13: *"The link looks odd"*). `.vz-c-name a` — the linked cell in every
   operator table — is `color: inherit; text-decoration: none` with the underline on
   `:hover`, and the owner's call on 2026-09-26 was *"underline on the portal only"*
   (`StylesheetReachTest` holds it). ▶ This row was the exception, not the rule.

   🔴 **THE `::after` IS WHAT MAKES THE WHOLE ROW THE TARGET**, which is the only way the
   arrow beside the FIGURE can be pressable while the link's accessible name stays the
   label's own WORDS. ⛔ The alternative was an `<a>` spanning `1 / -1` with its own copy of
   the `grid-template-columns` above — two statements of one layout in a rule seventeen
   widgets share, and the pair would drift on the first edit.

   ⚠ **NO `min-height: var(--vz-tap-min)` HERE, AND THAT IS A DECISION WITH A MEASUREMENT
   BEHIND IT.** `.vz-c-name a` takes the 44px, but `§7` item 20 priced the floor
   product-wide at **+8% page height** and the owner took it only on the members list. The
   overlay already gives the row's full width by its ~36px padding box, well past WCAG
   2.5.8 AA's 24×24. ⛔ And putting 44 on the LINK alone would make actionable rows taller
   than clear ones — F14's raggedness, arriving through the front door. */
.vz-widget-row-link {
  color: inherit;
  text-decoration: none;
}

.vz-widget-row-link::after {
  content: "";
  position: absolute;
  inset: 0;
}

.vz-widget-row-link:hover { text-decoration: underline; }

.vz-widget-row-link:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
  text-decoration: underline;
}

/* Column 2. `margin-left: auto` did this job under flex and is gone: in a grid it would be
   describing the placement twice, and the two could disagree after an edit. */
.vz-widget-row-value {
  grid-column: 2;
  justify-self: end;
  font-weight: var(--vz-fw-medium);
  color: var(--vz-ink);
  font-variant-numeric: tabular-nums lining-nums;
}

/* Its own line: a date and a state are context for the row above them, and
   wrapping them beside the amount reads as part of the figure.
   ⚠ `1 / -1` REPLACES `flex-basis: 100%` — same intent, stated in the grid's own terms. */
.vz-widget-row-meta {
  grid-column: 1 / -1;
  font-size: var(--vz-fs-xs);
  color: var(--vz-muted);
}

/* 🔴 ⚠ THE WIDGET'S HEADER, AND IT IS THE DRAG HANDLE (owner, 2026-09-07).
   `draggable` moved off the `<section>` and onto this row. That is not cosmetic:
   a draggable section swallows text selection over every figure inside it, and
   on a chart it fought the canvas for the pointer.

   ⚠ `cursor: grab` IS THE ONLY THING THAT SAYS SO. Nothing else on the page
   announces that a header can be dragged, and HTML5 drag has no affordance of
   its own. It is on the head and NOT on the title's link, which keeps a pointer
   over the drill-through reading as a link.

   ⚠ `align-items: flex-start` SO A TWO-LINE WIDGET NAME KEEPS ITS BUTTONS IN
   THE CORNER. Centring would drift them down the box as the heading wrapped,
   which at 320px is most of them. */
/* 🐛 ⚠ `flex-wrap` AND A REAL BASIS ON THE TITLE — THIS IS P5's RIBBON AGAIN,
   AND IT WAS MEASURED HERE BEFORE IT SHIPPED. Three 44px tap targets are 140px
   of a 228px KPI tile, so on the first render the title box came back **38px
   wide** carrying "Total members" — a two-word column beside its own controls,
   with `documentElement.scrollWidth - clientWidth` reading **0**, because
   nothing was overflowing. Exactly `.vz-widget-card`'s defect one component
   along, and exactly `.vz-facts`' before that.

   The basis is what makes the wrap happen: below ~9rem of available title
   column the buttons drop to their own line and the name takes the full width
   of the tile. Shrinking the buttons instead would have meant breaking
   `--vz-tap-min`, which is the floor this product holds everywhere. */
.vz-widget-head {
  display: flex;
  flex-wrap: wrap;
  align-items: flex-start;
  justify-content: space-between;
  gap: var(--vz-space-1) var(--vz-space-2);
  margin-bottom: var(--vz-space-3);
}

/* 🔴 ⚠ THE GRAB CURSOR IS KEYED OFF `[draggable]`, WHICH ONLY `dashboard.js` SETS
   (owner, 2026-09-07). Nothing else on the page announces that a header can be
   dragged, so this cursor IS the affordance — and an affordance on a phone, or with
   the script blocked, would promise a gesture that cannot fire. The attribute and the
   capability are therefore set together, in one place, and the pixels follow them. */
.vz-widget-head[draggable="true"] { cursor: grab; }
.vz-widget-head[draggable="true"]:active { cursor: grabbing; }

/* 🔴 ⚠ THE GRIP HANGS OFF THE SAME ATTRIBUTE THE CURSOR DOES, AND THAT IS THE
   CHANGE (owner, 2026-09-07). Until this row the cursor was the only thing that
   said a header could be dragged — and a cursor is invisible until the pointer is
   already on the right eight pixels, so the gesture was undiscoverable by anybody
   who had not been told about it.

   ⚠ ONE HOOK, NOT TWO CONDITIONS THAT COULD DISAGREE. `dashboard.js` sets
   `draggable` once, and only where a drag can actually fire — not on a coarse
   pointer, not with the script blocked. Everything visual keys off that same
   attribute, so there is no second rule anywhere that could be edited into
   promising a gesture the behaviour will not perform. `§8a`'s objection to a
   handle — *"an affordance advertising an interaction half the audience cannot
   perform is worse than none"* — is answered by the gate, not overruled.

   ⚠ AND IT IS `display: none` BY DEFAULT RATHER THAN REVEALED BY A CLASS. The
   markup ships it on every widget; the ONE thing that turns it on is the one thing
   that turns the drag on. A `.is-draggable` class would be a second source of
   truth for the same fact. */
/* 🐛 ⚠ AND IT TAKES THE SAME 44px FLOOR THE TITLE DOES, FOR THE SAME REASON AND
   AGAINST THE SAME WRONG FIRST ANSWER. `align-self: center` looks right on a
   one-line head and drops the grip **29px** down the box on a three-line one —
   measured at 769px on *"Active vs inactive members"*, where the ✕ stays in the
   corner and the grip slid to the middle. A floor centres the 16px glyph inside its
   own box at the top of the head instead, so the grip, the title's first line and
   the ✕ share one centre at every head height. ⚠ **It is a floor and not a tap
   target**: the grip is decoration, and the whole header is the drag surface. */
.vz-widget-grip {
  display: none;
  align-items: center;
  flex: 0 0 auto;
  min-height: var(--vz-tap-min);
  color: var(--vz-subtle);
}

.vz-widget-head[draggable="true"] .vz-widget-grip { display: inline-flex; }

/* ⚠ THE TITLE OWNED THE BOTTOM MARGIN UNTIL THE HEAD EXISTED. It is now inside
   a row that owns the spacing, and `min-width: 0` is what lets a long widget
   name wrap instead of pushing the buttons off the right edge — the rule
   `testEveryGrowableFlexItemEitherCanShrinkOrSaysWhyNot` exists for. */
/* 🔴 ⚠ THE BASIS WAS 9rem AND IT WAS MEASURED FOR THREE BUTTONS (§D12b). Row 5's
   corner held Move up, Move down and Remove — 140px of `--vz-tap-min` — and 9rem
   is what stopped a 228px KPI tile squeezing the title into a 38px column. The
   owner deleted two of those three the same day; the wrap threshold stayed where
   it was, so the ✕ went on dropping to its own line under a title that now had
   ~180px of room it was not being allowed to use. **The corner is 44px now, so
   the basis is the room 44px leaves.**

   ⚠ IT IS STILL A WRAP AND NOT A `nowrap`, because the head must still break at
   320px where a full-width widget's title genuinely is longer than its line.
   Measured through the `§E6` harness at every tile width the board can produce. */
/* 🐛 ⚠ AND `min-height` IS WHAT PUTS THE ✕ ON THE TITLE'S LINE RATHER THAN MERELY
   ON ITS ROW (§D12b). With the ✕ back beside the heading the head is at least
   `--vz-tap-min` tall — the button sets it — so a ONE-LINE title pinned to the top
   of a 44px row sat 10px above the ✕'s optical centre. **Measured at 375px, where
   every title is one line.** It reads as a ✕ floating below the name.

   ⚠ THE FIX IS ON THE TITLE, NOT ON `align-items`, AND THE DIFFERENCE IS THE THREE-
   LINE CASE. Centring the head would have aligned the one-line tile and pushed the
   ✕ **15px** down the box on *"Active vs inactive members"*, which is 74px tall at
   769px — worse than the defect, and exactly the drift `align-items: flex-start` was
   chosen to avoid. Giving the title a 44px floor instead makes a short heading centre
   inside its OWN box (`.vz-widget-title` is already `align-items: center`) and leaves
   a tall one untouched at the top. ⚠ It adds no height anywhere: the head was already
   44px, because the tap target is. */
.vz-widget-head .vz-widget-title {
  flex: 1 1 6rem;
  min-width: 0;
  min-height: var(--vz-tap-min);
  margin-bottom: 0;
}

/* The three controls. `flex: 0 0 auto` so they never shrink under a long
   heading — a half-width ✕ is a target nobody can hit — and `margin-inline-start:
   auto` keeps them hard right on the line they wrap onto, where
   `justify-content` no longer reaches them. */
.vz-widget-actions {
  display: flex;
  flex: 0 0 auto;
  align-items: center;
  gap: var(--vz-space-1);
  margin-inline-start: auto;
  /* The forms are only wrappers around one button each. */
  cursor: default;
}

.vz-widget-actions form { display: contents; }

/* ⚠ A FULL TAP TARGET, EVEN THOUGH IT IS AN ICON. `--vz-tap-min` is the floor
   this product holds everywhere, and three 16px glyphs in a corner is exactly
   where it would be tempting to break it.

   ⚠ AND `box-sizing: border-box` BECAUSE IT SIZES AND PADS — `testEveryRuleThat
   SizesAndPadsSaysWhichBoxItMeans`, which exists because `.vz-menu-item` shipped
   without it. */
.vz-widget-action {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  box-sizing: border-box;
  min-width: var(--vz-tap-min);
  min-height: var(--vz-tap-min);
  padding: 0;
  border: 1px solid transparent;
  border-radius: var(--vz-radius-md);
  background: transparent;
  color: var(--vz-muted);
  cursor: pointer;
}

.vz-widget-action:hover:not(:disabled) {
  border-color: var(--vz-border);
  background: var(--vz-surface-2);
  color: var(--vz-ink);
}

/* ⚠ DISABLED, NOT HIDDEN, ON THE FIRST AND LAST ROW (see the view) — so it has
   to LOOK unavailable rather than merely being inert. */
.vz-widget-action:disabled { opacity: 0.4; cursor: not-allowed; }

.vz-widget-action:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* Remove is the one destructive control in the corner, so it is the one that
   colours on hover. It is NOT red at rest: three widgets' worth of red ✕ down a
   board reads as an error state. */
.vz-widget-action--danger:hover:not(:disabled) {
  border-color: var(--vz-danger);
  background: transparent;
  color: var(--vz-danger);
}

/* ⚠ WHAT IS LEFT OF THE BOTTOM DISCLOSURE, AND IT IS NOT "ARRANGE" (owner,
   2026-09-07). The Arrange drawer is gone; a chart's period is not an
   arrangement, and folding a `<select>` into a row of icon buttons would be
   worse than a disclosure. Four widgets have one; the other twenty-five do not. */
.vz-widget-options { margin-top: var(--vz-space-3); }

.vz-widget-options > summary {
  cursor: pointer;
  font-size: var(--vz-fs-xs);
  color: var(--vz-muted);
}

/* ===========================================================================
   The widget panel (row 1, 2026-09-06; a right drawer, 2026-09-07)

   ⚠ IT IS A `<details>` STYLED AS A DRAWER, NOT A DIALOG. `§B-JS`'s rule is that
   an enhancement never becomes the only way to reach something, and a panel that
   only exists once script has opened it fails that. `<summary>` is focusable and
   toggles with no JavaScript, which is the same vocabulary the branch switcher
   and the phone drawer already use — a `<dialog>` would need `showModal()` and
   would therefore not exist at all with the script blocked.

   ⚠ AND `.vz-panel` IS A DIFFERENT COMPONENT. That one is the bordered block
   every module screen builds sections from; this is a disclosure. They were named
   apart deliberately, because inheriting a block's padding and border into a
   drawer is how one component quietly becomes two.
   =========================================================================== */
.vz-widget-panel { margin: 0; }

/* The summary IS the button, so it takes `.vz-action--primary`'s look from the
   markup and only loses the marker here. `list-style` covers Firefox and the
   `::-webkit-details-marker` rule covers Safari; neither alone is enough.

   🐛 `.vz-disclosure` IS THE SAME BEHAVIOUR WITHOUT THE DRAWER (F27 follow-up,
   owner 2026-09-24: *"Change billing and Record a new discount is not visible
   well"*). The billing panel's two money forms open from a `<summary>` that was
   rendering as bare text with a triangle, on a page where every other act is a
   `.vz-action` — and it is the ONLY route to two step-up writes, so a reader who
   does not read it as a control concludes the panel is read-only.
   ▶ ADDED TO THIS SELECTOR RATHER THAN GIVEN ITS OWN BLOCK, because it is the
   same four declarations for the same reason, and a second copy is how the two
   drift the next time somebody fixes the marker in one browser.
   ⚠ `.vz-widget-panel` KEEPS ITS OWN `margin: 0` above — that is about the drawer
   it anchors, not about the summary, which is why only this rule is shared. */
.vz-widget-panel > summary,
.vz-disclosure > summary {
  display: inline-flex;
  align-items: center;
  gap: var(--vz-space-2);
  cursor: pointer;
  list-style: none;
}

.vz-widget-panel > summary::-webkit-details-marker,
.vz-disclosure > summary::-webkit-details-marker { display: none; }

/* 🔴 ⚠ A DRAWER AT THE RIGHT EDGE, NOT A DROPDOWN UNDER THE BUTTON (owner,
   2026-09-07). Row 1 anchored the panel to the page header — a 26rem column
   pinned top-right above 60rem, an ordinary block beneath the button below it —
   and it read as a dropdown because that is what it was. It is now fixed to the
   viewport's right edge at full height, which is what "open a right-side panel"
   means and what a list somebody scrolls through wants.

   ⚠ IT OVERLAYS THE BOARD RATHER THAN PUSHING IT. Beyond being the conventional
   pattern, pushing would reflow the grid on every open — and since §D12a row 3
   the board holds canvases, so that is Chart.js resizing up to eight of them
   each time somebody opens the panel and again when they close it. Overlaying
   costs nothing.

   ⚠ AND IT IS STILL A `<details>`. §B-JS's rule is that an enhancement never
   becomes the only way to reach something: `<summary>` opens and closes this
   with no JavaScript, and the summary stays visible in the page header beside
   the drawer rather than behind it, so the no-JS close path is a real one. */
.vz-widget-panel-body {
  position: fixed;
  z-index: 30;
  inset-block: 0;
  inset-inline-end: 0;
  /* 🐛 ⚠ `max-width: 100%`, NEVER `min(26rem, 100vw)`. **`100vw` INCLUDES THE
     SCROLLBAR** — on a 320px viewport with a 15px scrollbar it resolves to 320px
     of drawer in 305px of space, so the box is pushed 15px off the LEFT edge and
     the first character of every widget name is cut off. Measured: `left: -15`
     at 320px and 375px, with `documentElement.scrollWidth - clientWidth` reading
     **0 the whole time**, because a fixed element overflowing does not scroll
     the page. `%` resolves against the containing block, which is the viewport
     the scrollbar has already been taken out of. */
  width: 26rem;
  max-width: 100%;
  /* ⚠ NOT `100vh`. On mobile Safari that is the window INCLUDING the browser
     chrome, so the foot of the drawer — which is where Reset lives — sits under
     the address bar. `100%` against a `fixed` box with `inset-block: 0` is the
     visual viewport and needs no unit trick. */
  height: 100%;
  overflow-y: auto;
  overscroll-behavior: contain;
  /* 🐛 ⚠ LOAD-BEARING, AND `StylesheetStaticTest` REFUSED THE RULE WITHOUT IT. The
     drawer is `width: min(26rem, 100vw)` WITH padding, so under the default
     `content-box` a 320px viewport gets 320px of box plus 32px of padding — the
     drawer overflowing the screen it is pinned to. This is E5b's `.vz-menu-item`
     defect in a second costume, and the test exists because of it. */
  box-sizing: border-box;
  padding: var(--vz-space-4);
  border-inline-start: 1px solid var(--vz-border);
  background: var(--vz-surface);
  box-shadow: var(--vz-shadow-md);
}

/* 🔴 ⚠ IT SLIDES IN FROM THE EDGE IT IS PINNED TO (owner, 2026-09-07). A drawer
   that simply appears reads as a repaint rather than as a panel opening, and on
   a board of figures the eye cannot tell what changed.

   ⚠ AN `animation` ON `[open]`, NOT A `transition`. A `<details>` body is not in
   the box tree at all while the element is closed, so there is no from-state for
   a transition to interpolate — it would fire on neither open nor close. A
   keyframe runs the moment the box exists, which is exactly the open.

   ⚠ THE CLOSE IS INSTANT, AND THAT IS HONEST RATHER THAN A GAP. Animating it
   out would mean holding the drawer in the DOM after `open` is gone, which needs
   script — and `§B-JS` fixes that this panel works with the script blocked.
   E13's rule is that motion is a courtesy, never a mechanism.

   ⚠ AND IT RESPECTS `prefers-reduced-motion`, which is not optional: a 26rem box
   flying across the viewport is precisely the movement that setting is for. */
@keyframes vz-widget-panel-in {
  from { opacity: 0; transform: translateX(1.5rem); }
  to   { opacity: 1; transform: translateX(0); }
}

.vz-widget-panel[open] > .vz-widget-panel-body {
  animation: vz-widget-panel-in 180ms ease-out;
}

@media (prefers-reduced-motion: reduce) {
  .vz-widget-panel[open] > .vz-widget-panel-body { animation: none; }
}

/* The drawer's own header — the title and the control that dismisses it. */
.vz-widget-panel-head {
  display: flex;
  align-items: flex-start;
  justify-content: space-between;
  gap: var(--vz-space-3);
}

/* ⚠ HIDDEN UNTIL `dashboard.js` REVEALS IT, like the search row above. Closing a
   `<details>` from inside itself needs script, so without that script this
   button would be a control that does nothing — and the summary in the page
   header is the working close path. */
.vz-widget-panel-close[hidden] { display: none; }

.vz-widget-panel-close {
  flex: 0 0 auto;
  min-width: var(--vz-tap-min);
  min-height: var(--vz-tap-min);
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  color: var(--vz-muted);
  cursor: pointer;
}

.vz-widget-panel-filters {
  display: flex;
  flex-direction: column;
  gap: var(--vz-space-2);
  margin: var(--vz-space-3) 0;
}

/* ⚠ EXPLICIT, BECAUSE `display` ON THE CLASS BEATS THE UA'S `[hidden]`. This
   product has no global `[hidden] { display: none }` rule — only per-component
   ones, which is why each element `dashboard.js` toggles needs its own. Without
   this line the search row is revealed on every page load, script or not, and the
   `hidden` attribute in the markup does nothing at all. */
.vz-widget-panel-filters[hidden] { display: none; }

.vz-widget-panel-foot {
  margin-top: var(--vz-space-4);
  padding-top: var(--vz-space-3);
  border-top: 1px solid var(--vz-border);
}

/* ===========================================================================
   The KPI tiles (§D12a row 2)

   ⚠ A TILE IS THE SAME `.vz-widget` SECTION AS A PANEL, at a quarter width and
   with a bigger figure. It is NOT a second component: the border, the padding,
   the heading and the arrange disclosure are all the board's, so a tile and a
   panel cannot drift apart in anything but the size of the number.
   =========================================================================== */

/* ⚠ NOTHING RESIZES `.vz-widget-figure` HERE, AND THAT IS DELIBERATE. It is
   already `--vz-fs-display` with `tabular-nums lining-nums` — a headline figure
   is what D12 built it for, and a tile is that same headline in a narrower box.
   A second rule setting `font-size` would also have dropped `lining-nums` by
   restating `font-variant-numeric` incompletely, which is how four tiles end up
   with digits that do not line up in a row.

   ⚠ AND NO `overflow-wrap: anywhere` ON IT EITHER. Row 14 recorded what that does
   to a table column — min-content becomes ONE CHARACTER, and the box gets crushed
   to a vertical stack of letters at some width in the middle of the range, at zero
   page overflow the whole time. A long money figure is handled by the tile being a
   quarter of twelve columns, not by letting it break anywhere. */

/* ⚠ THE HEIGHT IS ON THE WRAPPER, NEVER ON THE `<canvas>` (§D12a row 3). With
   `maintainAspectRatio: false` Chart.js sizes itself to its parent and rewrites
   the element's own width/height attributes on every resize — a height set on
   the canvas is overwritten and the chart collapses to nothing. `position:
   relative` is what gives it a box to measure. */
/* ⚠ DRAG FEEDBACK, AND NOTHING ELSE (§D12a row 5). A dragged widget dims and the
   one under the pointer takes a rule on the edge the drop lands against. No
   placeholder gap and no animated reflow: the board is a flow, so what a drop
   produces is simply the next render — and a preview that guessed differently
   would be a second layout engine disagreeing with the server. */
.vz-widget.is-dragging { opacity: 0.45; }

.vz-widget.is-drop-target {
  outline: 2px solid var(--vz-primary);
  outline-offset: -2px;
}

/* 🐛 ⚠ THIS BLOCK USED TO SAY *"THE CURSOR IS THE ONLY AFFORDANCE, AND THAT IS
   DELIBERATE"* AND TO KEY ITSELF OFF `.vz-widget[draggable="true"]` — a selector
   that has matched NOTHING since the handle moved to `.vz-widget-head` on
   2026-09-07. Both of its sentences had also stopped being true: it argued from
   *"the up/down buttons are the path"* for buttons the owner deleted the same day,
   and §D12b has now drawn the handle it says should not exist. A rule that cannot
   match, under a comment arguing for a decision that was reversed, is exactly the
   thing `§D11d` records a whole derivation being spent on. The cursor rules live
   on the head, where the attribute is; only the dragging cursor is kept here,
   because `is-dragging` is a class on the SECTION.

   ⚠ IT STAYS INSIDE THE POINTER QUERY. `dashboard.js` already refuses to bind on a
   coarse pointer, so this is belt and braces — but it is free, and it keeps the
   stylesheet's own answer to *"can this device drag"* in one shape. */
@media (hover: hover) and (pointer: fine) {
  .vz-widget.is-dragging { cursor: grabbing; }
}

/* ===========================================================================
   The KPI tile takes Finance's HIERARCHY, and the reason is a measurement
   (§D12b)

   ── 🔴 WHAT WAS TRIED FIRST, AND WHY THE NUMBERS REFUSED IT ────────────────
   `vazhai-finance`'s `.vz-kpi__icon` is a tinted rounded box beside the figure, and
   putting it on this board was the obvious reading of *"KPI restyling from Finance's
   CSS"*. It does not fit, at any size, and the arithmetic is worth keeping because
   the next reader will try it too.

   A quarter tile is **228px** at 1440 with the sidebar out, so **188px** of content.
   The head must carry the grip (16), the ✕ (`--vz-tap-min`, 44) and three 8px gaps —
   **68px of chrome** — and the longest tile name, *"Value under management"*,
   measures **97px at min-content** in `--vz-fs-md`. 16+8+18+8+97+8+44 = **199 > 188**
   with only the plain label glyph, which is why the shipped board could afford it:
   the ✕ was WRAPPING to its own line and handing the title the full 186px. Putting
   the ✕ back on that row is this row's instruction, so the row got 76px tighter and
   a 36px tinted box needs 44 more.

   Beside the FIGURE instead — which is where Finance actually puts it — is worse:
   `₹25,00,00,000.00` at `--vz-fs-display` is **168px of the same 188px**, and
   `§D12a` row 2 already fixed a money figure clipped inside a tile. A badge there
   re-creates that defect, and shrinking the figure was refused by that row too.

   ── ✅ SO THE PART OF FINANCE'S KPI THAT SHIPS IS ITS TYPE HIERARCHY ────────
   `.vz-kpi__label` is `--vz-fs-sm` and secondary; `.vz-kpi__value` is
   `--vz-fs-display`. Chit's tile shouted BOTH — a 16px bold link over a 22px bold
   number — so the thing the tile exists to say competed with its own caption. This
   is not a consolation prize for the badge: it is the rule that makes a KPI card
   read as one figure, and it is the half that fits.

   ⚠ AND IT BUYS THE ROOM BACK, WHICH IS THE CONSTRAINT AND THE DESIGN AGREEING.
   *"Value under management"* falls from 97px to ~79px of min-content at 13px, so
   16+8+18+8+79+8+44 = **181 ≤ 188** and the ✕ stays on the title's row.

   ⚠ THE LINK'S COLOUR DOES NOT MOVE. Finance's label is `--vz-text-secondary`
   because the whole CARD is the anchor there; here the heading IS the drill-through
   and `--vz-primary-strong` is what says so on every widget in the product. Size and
   weight are the hierarchy; colour is the affordance, and only one of them is
   Finance's to lend. */
.vz-widget--tile .vz-widget-title {
  font-size: var(--vz-fs-sm);
  font-weight: var(--vz-fw-medium);
}

.vz-widget-config {
  display: flex;
  flex-wrap: wrap;
  align-items: end;
  gap: var(--vz-space-2);
  margin-top: var(--vz-space-3);
  padding-top: var(--vz-space-3);
  border-top: 1px solid var(--vz-border);
}

/* The select takes the room it needs rather than the block width `.vz-control`
   gives it inside a form grid. */
.vz-widget-config .vz-control { width: auto; min-width: 9rem; }

/* ===========================================================================
   The auction calendar and the action queue (§D12a row 4)

   ⚠ A LIST GROUPED BY DAY, NOT A MONTH GRID. Seven columns at 375px gives each
   day about 40px, which is not a date and a group code — the same argument
   `PROJECT_PLAN §8a` used to refuse coordinates for the board itself. An
   operator's events are few and dated; a grid would be mostly empty cells drawn
   to look like a calendar.
   =========================================================================== */
/* 🔴 **A RULED LIST, AND THE WEIGHT IS ON THE DATE** (owner, 2026-10-13: *"can we revert the
   card style for auction calendar and add badge design for date?"*).

   🐛 **THE CARDS WERE MY ANSWER TO *"it looks bit un-visible"* AND THEY ANSWERED THE WRONG
   HALF.** The diagnosis was right — the date carried no more weight than the chit names under
   it — but a surface per day made the LIST heavier without making the DATE easier to find, and
   this widget already sits on a card, so a card per row was a card inside a card. ▶ The fix
   belongs on the date, which is the pill below.

   ⚠ THE VALUES HERE ARE THE ONES THIS BLOCK HELD BEFORE `f2e1865`, read back from git rather
   than remembered — the `:first-child` exception included, which exists to undo the `border-top`
   on the row that has nothing above it. */
.vz-calendar {
  margin: 0;
  padding: 0;
  list-style: none;
}

.vz-calendar-day {
  padding: var(--vz-space-3) 0;
  border-top: 1px solid var(--vz-border);
}

.vz-calendar-day:first-child { border-top: 0; }

/* ⚠ **A RULE ON THE INLINE EDGE, NOT A BACKGROUND WASH**, and this argument has now survived
   both designs: today's day sits among others a reader is scanning, a tinted block would be the
   loudest thing in the widget, and the badge beside the date already says the word.

   ⚠ AND THE `padding-inline-start` IS BACK TO A WHOLE STEP. It was `calc(space-3 - 2px)` only
   because the card had a 1px border of its own to stay flush with; with no card there is nothing
   to subtract from, and without the padding the date would hug the rule. */
.vz-calendar-day.is-today {
  border-inline-start: 3px solid var(--vz-primary);
  padding-inline-start: var(--vz-space-3);
}

/* 🔴 **THE DATE IS A BADGE** (owner, 2026-10-13: *"add badge design for date"*). It is the thing
   a reader scans this widget for, and a pill gives it an edge the surrounding text has not —
   which is what the row cards were reaching for, applied to the element that actually carries
   the question.

   ⚠ **THE PILL IS ON THE `<time>`, NOT ON THE `<p>`**, and that is both why the view needs no
   change and why the *Today* badge beside it stays a separate object. Two pills side by side
   read as two facts; one pill wrapped around both would read as one — and on eleven of the
   twelve rows the second fact is not there at all.

   ⚠ `--vz-surface-2` AND `--vz-ink`, NOT A TINT. Every date is the same kind of thing, so the
   tone would carry no meaning, and a coloured pill would compete with the `.vz-badge--*` states
   to its right, which DO mean something. ⚠ `--vz-fs-sm` and not a step up: the badge's own edge
   is what makes it findable now, and a 16px bold pill beside a 13px *Today* badge would read as
   two different kinds of label. `tabular-nums` so a column of dates does not jitter as the
   digits change width. */
.vz-calendar-date {
  display: flex;
  align-items: center;
  gap: var(--vz-space-2);
  margin: 0 0 var(--vz-space-2);
}

.vz-calendar-date time {
  display: inline-flex;
  align-items: center;
  padding: var(--vz-space-1) var(--vz-space-2);
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface-2);
  color: var(--vz-ink);
  font-size: var(--vz-fs-sm);
  font-weight: var(--vz-fw-bold);
  font-variant-numeric: tabular-nums;
}

.vz-calendar-events {
  margin: 0;
  padding: 0;
  list-style: none;
}

.vz-calendar-events li + li { margin-top: var(--vz-space-1); }

/* ⚠ THE WHOLE ROW IS THE TARGET, NOT THE GROUP CODE INSIDE IT. `§7` item 19's
   floor is about what a thumb can hit, and a 44px band beats a 60px word. */
.vz-calendar-event {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-2);
  min-height: var(--vz-tap-min);
  padding: var(--vz-space-1) var(--vz-space-2);
  border-radius: var(--vz-radius-md);
  color: var(--vz-ink);
  text-decoration: none;
}

.vz-calendar-event:hover { background: var(--vz-surface-2); }
.vz-calendar-event:hover .vz-calendar-event-name { text-decoration: underline; }

.vz-calendar-event-name {
  flex: 1 1 auto;
  min-width: 0;
  font-weight: var(--vz-fw-medium);
  color: var(--vz-primary-strong);
}

.vz-calendar-event-meta {
  flex: 0 1 auto;
  min-width: 0;
  font-size: var(--vz-fs-sm);
  color: var(--vz-muted);
}

/* ===========================================================================
   The month calendar — /app/auctions/calendar (owner, 2026-10-13)
   ---------------------------------------------------------------------------
   🔴 **ONE MARKUP, TWO LAYOUTS, AND THAT IS THE WHOLE DESIGN.** The owner asked for the
   seven-column grid in their reference screenshot; a seven-column grid at 375px gives each
   day about 40px, which is not a date and a group code — the argument the dashboard widget
   above and `PROJECT_PLAN §8a` both refuse a grid on. ▶ The cells are a FLAT list, the grid
   is `repeat(7, …)`, and below 48rem the columns collapse to one and the empty days drop
   out. What is left is the day-grouped list the widget already draws.

   ⛔ **RENDERING BOTH AND HIDING ONE WOULD PUT A MONTH OF EVENTS ON THE WIRE TWICE**, on a
   product whose compression is HELD (`F22`), so raw bytes are wire bytes.
   =========================================================================== */
/* 🐛 **THE RIGHT AND BOTTOM EDGES ARE DRAWN BY THE CELLS, NOT BY THIS BOX — MEASURED.**
   With a border on all four sides the rightmost cell’s inset rule landed 1px inside the
   container’s border and the two read as a single 2px line, against 1px everywhere else:
   the container ended at 1220 and the last cell at 1219. ▶ The grid always fills seven
   columns and whole weeks, so the cells’ own rules close the box exactly. */
.vz-month {
  margin-top: var(--vz-space-4);
  border: 1px solid var(--vz-border);
  border-right-width: 0;
  border-bottom-width: 0;
  border-radius: var(--vz-radius-md);
  overflow: hidden;
  /* A white sheet on the grey page (owner, 2026-11-19: "it looks plain"). */
  background: var(--vz-surface);
}

/* ⚠ **ONE RULE, TWO SELECTORS — THE HEADER STRIP AND THE CELLS CANNOT BE GIVEN DIFFERENT
   TRACKS.** A header one column out of step with the grid beneath it is wrong in a way that
   reads as right, and it is the single most likely thing to go wrong on this screen. The
   ORDER of the headings is settled in the same place, by `AuctionCalendar::weekdayKeys()`,
   which derives them from `WEEK_STARTS_ON` rather than listing them. */
.vz-month-head,
.vz-month-grid {
  display: grid;
  grid-template-columns: repeat(7, minmax(0, 1fr));
  margin: 0;
  padding: 0;
  list-style: none;
}

.vz-month-head {
  background: var(--vz-surface-2);
  border-bottom: 1px solid var(--vz-border);
}

.vz-month-head li {
  padding: var(--vz-space-2);
  font-size: var(--vz-fs-xs);
  font-weight: var(--vz-fw-medium);
  color: var(--vz-muted);
  text-align: center;
}

/* ⚠ **THE RULES ARE DRAWN AS INSET SHADOWS, NOT BORDERS.** Seven columns of `border-right`
   leave a doubled 2px line wherever two cells meet, and the usual fix — a negative margin —
   fights the grid's own track sizing. An inset shadow paints inside the cell's box and
   costs the layout nothing.

   ⚠ AND `min-height` RATHER THAN A FIXED ONE: a day with four draws must grow, and a row of
   a calendar is as tall as its fullest cell whatever this says. */
.vz-month-day {
  box-sizing: border-box;
  min-height: 5.5rem;
  padding: var(--vz-space-2);
  box-shadow:
    inset -1px 0 0 var(--vz-border),
    inset 0 -1px 0 var(--vz-border);
}

/* The month's own days are the subject; its neighbours' are context. ⚠ A TINT RATHER THAN
   MUTED INK, because the DATE inside still has to be readable — these cells carry real
   draws, and a draw on the 1st of next month reached from this month's last row is the
   reason the padding holds real dates at all. */
/* ⚠ (2026-11-19) WHITE LIKE THE REST, WITH FADED NUMBERS — the grey tint now means a BUSY day,
   and two meanings for one shade would make next month's 1st look like a full day. */
.vz-month-day.is-outside .vz-month-event { opacity: 0.7; }

/* ⚠ THE SAME DECISION THE WIDGET MAKES, FOR THE SAME REASON: a rule on the inline edge and
   not a background wash. Today's cell sits among forty-one others a reader is scanning; a
   tinted block would be the loudest thing on the page, and the badge beside the number
   already says the word. ⚠ THE SHADOW IS RESTATED rather than added to, because
   `box-shadow` does not cascade by part — dropping the two inset rules here would take this
   one cell out of the grid's ruling. */
.vz-month-day.is-today {
  box-shadow:
    inset 3px 0 0 var(--vz-primary),
    inset -1px 0 0 var(--vz-border),
    inset 0 -1px 0 var(--vz-border);
}

.vz-month-date {
  display: flex;
  align-items: center;
  gap: var(--vz-space-1);
  margin: 0 0 var(--vz-space-1);
}

.vz-month-num {
  font-size: var(--vz-fs-sm);
  font-weight: var(--vz-fw-bold);
  font-variant-numeric: tabular-nums;
  color: var(--vz-ink);
}

.vz-month-day.is-outside .vz-month-num { color: var(--vz-muted); }

.vz-month-events {
  display: flex;
  flex-direction: column;
  gap: 2px;
  margin: 0;
  padding: 0;
  list-style: none;
}

/* ⚠ **THE CHIP IS THE TARGET AND IT WRAPS.** A group code and an installment number is 12–16
   characters in a column that is about 150px at 1280 and narrower on a tablet, so the name
   and the badge have to be allowed onto a second line rather than overflowing the cell —
   which is the one thing a calendar grid cannot do without pushing the page sideways. */
.vz-month-event {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-1);
  padding: 2px var(--vz-space-1);
  border-radius: var(--vz-radius-sm);
  background: var(--vz-primary-tint);
  color: var(--vz-primary-strong);
  font-size: var(--vz-fs-xs);
  text-decoration: none;
}

.vz-month-event:hover { box-shadow: inset 0 0 0 1px currentColor; }
.vz-month-event:hover .vz-month-event-name { text-decoration: underline; }

.vz-month-event:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* A held draw is history and an upcoming one is work. ⚠ THE WORD *Settled* IS BESIDE IT AND
   CARRIES THE SAME FACT — this is the quieter of two signals, never the only one. */
/* (2026-11-19) To be held is a green chip; held is a white one with a rule round it, so it
   still reads on a shaded busy day. */
.vz-month-event.is-settled {
  background: var(--vz-surface);
  box-shadow: inset 0 0 0 1px var(--vz-border);
  color: var(--vz-muted);
}

.vz-month-event-name {
  min-width: 0;
  font-weight: var(--vz-fw-medium);
}

.vz-month-nav {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-2);
  margin-top: var(--vz-space-4);
}

/* ===========================================================================
   Day, week and month (owner, 2026-11-19 — from the preview they approved)
   ---------------------------------------------------------------------------
   🔴 EACH VIEW IS DRAWN ONCE AND RE-LAID BELOW 48rem, never drawn twice and hidden (`F22`).
   =========================================================================== */

/* The time leads the chip, because in a month or a week WHEN is the question being asked. */
.vz-month-event-time {
  font-variant-numeric: tabular-nums;
  color: var(--vz-muted);
}

.vz-month-event-flag { width: 14px; height: 14px; flex: none; vertical-align: -2px; }

/* Each piece stays whole; the chip wraps BETWEEN them. */
.vz-month-event-code { white-space: nowrap; }
.vz-month-event-time { display: inline-flex; align-items: center; gap: 2px; }

/* ⚠ TWO AUCTIONS CLOSER THAN `AuctionClash::MINUTES`. Amber, with the sentence in a
   `.vz-sr-only` beside it and in the legend below — never the colour alone. */
.vz-month-event.is-clash,
.vz-day-card.is-clash {
  background: var(--vz-warning-tint);
  color: var(--vz-warning-text);
}

.vz-month-event.is-clash .vz-month-event-time { color: inherit; }

/* A day with `BUSY_DAY` or more auctions. A tint, like `is-outside`, so the chips stay legible. */
.vz-month-day.is-busy { background: var(--vz-surface-2); }

.vz-month-more {
  display: inline-block;
  padding: 2px var(--vz-space-1);
  color: var(--vz-muted);
  font-size: var(--vz-fs-xs);
}

/* The phone's whole cell — hidden where the chips say it at length. */
.vz-month-count { display: none; }

.vz-cal-bar {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  justify-content: space-between;
  gap: var(--vz-space-2);
  margin-top: var(--vz-space-4);
}

.vz-cal-bar .vz-month-nav { margin-top: 0; }

/* A segmented control, built from links: each view is a read with its own URL. */
.vz-cal-views {
  display: inline-flex;
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  overflow: hidden;
}

.vz-cal-view {
  display: inline-flex;
  align-items: center;
  box-sizing: border-box;
  min-height: var(--vz-tap-min);
  padding: 0 var(--vz-space-4);
  color: var(--vz-ink);
  font-size: var(--vz-fs-sm);
  text-decoration: none;
}

.vz-cal-view + .vz-cal-view { border-left: 1px solid var(--vz-border); }
.vz-cal-view:hover { background: var(--vz-surface-2); }

.vz-cal-view[aria-current="page"] {
  background: var(--vz-primary-tint);
  color: var(--vz-primary-strong);
  font-weight: var(--vz-fw-medium);
}

.vz-cal-view:focus-visible { outline: 2px solid var(--vz-primary); outline-offset: -2px; }

.vz-cal-legend {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2) var(--vz-space-4);
  margin: var(--vz-space-3) 0 0;
  padding: 0;
  list-style: none;
  color: var(--vz-muted);
  font-size: var(--vz-fs-xs);
}

.vz-cal-legend li { display: inline-flex; align-items: center; gap: var(--vz-space-1); }

.vz-cal-key {
  display: inline-block;
  width: 12px;
  height: 12px;
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-sm);
  background: var(--vz-primary-tint);
}

.vz-cal-key.is-settled { background: var(--vz-surface); }
.vz-cal-key.is-clash { background: var(--vz-warning-tint); }
.vz-cal-key.is-busy { background: var(--vz-surface-2); }

.vz-cal-hint { display: none; }

/* The free half-hours offered under a same-time warning on the move form. */
.vz-clash-free {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-2);
  margin-top: var(--vz-space-3);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

.vz-clash-free p { margin: 0; }

/* ── THE WEEK ─────────────────────────────────────────────────────────────
   ⚠ DAY-MAJOR MARKUP ON A SHARED ROW GRID. Each day column is a `subgrid` over the outer
   rows, so 10 am lines up across all seven days however many auctions one slot holds — and
   the same markup stacks into a per-day list on a phone, which hour-major markup could not.
   ⚠ THIRTEEN ROWS ARE DECLARED (the heading, `early`, nine-to-six, `late`); a week that draws
   fewer leaves the last ones empty at no height. */
.vz-week {
  display: grid;
  grid-template-columns: 4.5rem repeat(7, minmax(0, 1fr));
  grid-template-rows: repeat(13, auto);
  margin-top: var(--vz-space-4);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  overflow: hidden;
  background: var(--vz-surface);
}

.vz-week-times,
.vz-week-day {
  display: grid;
  grid-row: 1 / -1;
  grid-template-rows: subgrid;
  min-width: 0;
}

.vz-week-day { border-left: 1px solid var(--vz-border); }
.vz-week-day.is-today { box-shadow: inset 0 3px 0 var(--vz-primary); }

.vz-week-corner,
.vz-week-head {
  margin: 0;
  padding: var(--vz-space-2);
  background: var(--vz-surface-2);
  border-bottom: 1px solid var(--vz-border);
  font-size: var(--vz-fs-xs);
  font-weight: var(--vz-fw-medium);
}

.vz-week-head { display: flex; flex-wrap: wrap; align-items: center; gap: var(--vz-space-1); }
.vz-week-head a { color: var(--vz-ink); }

.vz-week-time {
  padding: var(--vz-space-1) var(--vz-space-2);
  border-bottom: 1px solid var(--vz-border);
  color: var(--vz-muted);
  font-size: var(--vz-fs-xs);
  font-variant-numeric: tabular-nums;
}

.vz-week-slot {
  display: flex;
  flex-wrap: wrap;
  align-content: flex-start;
  gap: 2px;
  min-height: 2.25rem;
  padding: 2px;
  border-bottom: 1px solid var(--vz-border);
}

/* Two auctions in one hour sit side by side and share the column. */
.vz-week-slot .vz-month-event { flex: 1 1 4.5rem; min-width: 0; }

.vz-week-none { display: none; }

/* ── THE DAY ──────────────────────────────────────────────────────────────── */
.vz-day {
  margin: var(--vz-space-4) 0 0;
  padding: 0;
  list-style: none;
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  overflow: hidden;
  background: var(--vz-surface);
}

.vz-day-slot {
  display: grid;
  grid-template-columns: 6rem minmax(0, 1fr);
  border-bottom: 1px solid var(--vz-border);
}

.vz-day-slot:last-child { border-bottom: 0; }

.vz-day-time {
  padding: var(--vz-space-3) var(--vz-space-2);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
  font-variant-numeric: tabular-nums;
}

.vz-day-events {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2);
  min-width: 0;
  padding: var(--vz-space-2);
}

.vz-day-free {
  align-self: center;
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

/* Two auctions in one hour sit side by side, and wrap rather than squeeze. */
.vz-day-card {
  display: flex;
  flex: 1 1 14rem;
  flex-direction: column;
  gap: 2px;
  min-width: 0;
  padding: var(--vz-space-2) var(--vz-space-3);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  background: var(--vz-primary-tint);
  color: var(--vz-primary-strong);
  text-decoration: none;
}

.vz-day-card:hover .vz-day-card-title { text-decoration: underline; }
.vz-day-card:focus-visible { outline: 2px solid var(--vz-primary); outline-offset: 2px; }
.vz-day-card.is-settled { background: var(--vz-surface); color: var(--vz-muted); }

/* (2026-11-19) The whole day is drawn, so the hours outside nine-to-six sit quieter. */
.vz-day-slot.is-quiet .vz-day-time,
.vz-day-slot.is-quiet .vz-day-events { padding-top: var(--vz-space-1); padding-bottom: var(--vz-space-1); }
.vz-day-slot.is-quiet { background: var(--vz-surface-2); }
.vz-day-card.is-clash { border-color: var(--vz-warning); }

.vz-day-card-title { font-weight: var(--vz-fw-medium); overflow-wrap: anywhere; }
.vz-day-card-name { color: var(--vz-ink); font-size: var(--vz-fs-sm); overflow-wrap: anywhere; }
.vz-day-card-meta { color: var(--vz-muted); font-size: var(--vz-fs-sm); }

.vz-day-card-clash {
  display: inline-flex;
  align-items: center;
  gap: var(--vz-space-1);
  font-size: var(--vz-fs-sm);
}

/* 🔴 BELOW 48rem — the preview the owner approved on 2026-11-19.
   ▶ THE MONTH keeps its seven columns, and each day shows only a COUNT that opens that day:
   at 375px a column is about 48px, which holds a date and a number but not a group code.
   ▶ THE WEEK's day columns stack into a list of days, each naming its own auctions.
   ▶ THE DAY is one column already; its time column narrows. */
@media (max-width: 47.99rem) {
  .vz-month-day {
    min-height: 3.5rem;
    padding: var(--vz-space-1);
    text-align: center;
  }

  .vz-month-date { justify-content: center; margin: 0; }
  .vz-month-today { display: none; }
  .vz-month-events { display: none; }

  .vz-month-count {
    display: flex;
    align-items: center;
    justify-content: center;
    box-sizing: border-box;
    min-height: 1.75rem;
    margin-top: 2px;
    border-radius: var(--vz-radius-pill);
    background: var(--vz-primary-tint);
    color: var(--vz-primary-strong);
    font-size: var(--vz-fs-xs);
    font-weight: var(--vz-fw-bold);
    text-decoration: none;
  }

  /* ⚠ THE PRIMARY BUTTON'S MEASURED PAIR (`ThemeContrastTest`), not a surface token as ink. */
  .vz-month-day.is-busy .vz-month-count { background: var(--vz-primary-button); color: var(--vz-primary-contrast); }

  .vz-cal-hint {
    display: block;
    margin: var(--vz-space-2) 0 0;
    color: var(--vz-muted);
    font-size: var(--vz-fs-sm);
  }

  .vz-cal-legend li:nth-child(-n+3) { display: none; }

  .vz-week { display: block; }
  .vz-week-times { display: none; }

  .vz-week-day { display: block; border-left: 0; border-bottom: 1px solid var(--vz-border); }
  .vz-week-day:last-child { border-bottom: 0; }
  .vz-week-day.is-today { box-shadow: inset 3px 0 0 var(--vz-primary); }

  .vz-week-head { border-bottom: 0; font-size: var(--vz-fs-sm); }
  .vz-week-slot { min-height: 0; padding: 0 var(--vz-space-2) var(--vz-space-1); border-bottom: 0; }
  .vz-week-slot.is-empty { display: none; }
  .vz-week-slot .vz-month-event { flex-basis: 100%; font-size: var(--vz-fs-sm); }

  .vz-week-none {
    display: block;
    margin: 0;
    padding: 0 var(--vz-space-2) var(--vz-space-2);
    color: var(--vz-muted);
    font-size: var(--vz-fs-sm);
  }

  .vz-day-slot { grid-template-columns: 4.5rem minmax(0, 1fr); }
  .vz-day-card { flex-basis: 100%; }
}

/* ⚠ URGENCY, AND ONLY ON A COUNT THAT IS NOT ZERO — the view withholds the class
   otherwise, because a red `0` is a contradiction and would be the loudest thing
   on a clear board. ⚠ AND IT IS NEVER THE ONLY SIGNAL: every line names what it
   counts in words beside the figure. */
.vz-tone--urgent { color: var(--vz-danger-text); }
.vz-tone--soon   { color: var(--vz-warning-text); }

.vz-chart {
  position: relative;
  height: 15rem;
  margin-bottom: var(--vz-space-2);
}

/* A full-width chart has more to say and room to say it — twelve months of two
   series is unreadable at the height a doughnut needs. */
.vz-widget--full .vz-chart { height: 19rem; }

.vz-trend {
  display: flex;
  align-items: center;
  gap: var(--vz-space-1);
  margin: var(--vz-space-1) 0 0;
  font-size: var(--vz-fs-sm);
  font-variant-numeric: tabular-nums;
}

/* 🔴 ⚠ DIRECTION, NOT JUDGEMENT — AND THAT IS WHY NEITHER IS GREEN OR RED.
   Overdue payments rising is a bad month and members rising is a good one, so a
   stylesheet that painted `up` as success would be congratulating an operator on
   the one figure they need to worry about. The arrow character carries the
   direction, `.vz-sr-only` carries its name, and the colour carries only
   emphasis. */
.vz-trend--up,
.vz-trend--down { color: var(--vz-ink); }
.vz-trend--flat { color: var(--vz-muted); }

.vz-trend-since { color: var(--vz-subtle); }

/* 🐛 THE ROW HAD NO SPACE BENEATH IT ANYWHERE (owner, 2026-09-23, from a screenshot of
   `/admin/messaging`). `margin: 0` and a measured **0px gap to the next block on every
   screen that uses it** — so a chip's rounded bottom sat flush against the border of
   whatever followed, and on messaging that is a card, which reads as a collision rather
   than as tight spacing.
   ⚠ IT LOOKED ACCEPTABLE ON SOME SCREENS AND NOT OTHERS, WHICH IS WHY IT SURVIVED: where
   the next block is a `.vz-toolbar` the padding inside it hides the missing gap, and where
   it is a `.vz-table-wrap` the border lands on the pill. The bug was constant; only its
   visibility varied.
   ⛔ AND IT IS NOT A REGRESSION FROM THE CHIP RESTYLE — `git show HEAD` has the same
   `margin: 0`. Reducing the pills from 62px to 44px changed where the seam falls, not
   whether there is one. */
.vz-chips {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2);
  margin-bottom: var(--vz-space-4);
}

/* ⚠ 44px OF TARGET, NOT 44px OF INK. `§7` item 19's floor is about what a thumb
   can hit, so the padding does the work and the label stays small.

   🐛 AND IT WAS RENDERING AT 62px, NOT 44 (measured 2026-09-23, owner's item 4).
   `min-height` resolves against the CONTENT box by default, so the 8px padding and
   the 1px border were ADDED to the floor rather than counted inside it: 44 + 16 + 2.
   The result was a tall pill holding one 20px line of text, which is most of why the
   strip read as unfinished. ▶ `box-sizing: border-box` makes the number mean what the
   comment above always said it meant. ⚠ There is no global reset in this product —
   `border-box` is set per component, six places before this one — so this is the
   house convention rather than a new rule.

   🔴 AND IT IS A CONTROL, NOT PROSE, SO THE UNDERLINE GOES. Every chip is an `<a>`
   that filters the list, and the default underline made a row of pills look like a
   paragraph of links that had lost its styling. The affordance it carried is replaced
   rather than dropped: a border and background that answer on hover, plus the pill,
   the pointer and `aria-current` on the active one. */
.vz-chip {
  box-sizing: border-box;
  display: inline-flex;
  align-items: center;
  gap: var(--vz-space-1);
  min-height: 2.75rem;
  padding: var(--vz-space-2) var(--vz-space-3);
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface);
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
  line-height: 1.25;
  text-decoration: none;
  white-space: nowrap;
}

/* 🐛🔴 A CHIP IS NOT ALWAYS A CONTROL, AND THE BASE RULE USED TO CLAIM IT WAS.
   `cursor: pointer` sat on `.vz-chip` itself, so every chip showed a hand — including the
   ones that are inert. Measured across the views: **8 screens render `<a class="vz-chip">`**
   (announcements, audit, connections, errors, jobs, messaging/outbox, organizations,
   provisioning) and **2 render `<span class="vz-chip">`** — `/admin/migrations` and
   `/admin/messaging`, where the chips are a read-out and filter nothing.

   ⚠ **`/admin/messaging` LEFT THAT SECOND GROUP ON 2026-10-13** (owner: *"when clicked the
   status cards filter the template"*), so it now renders BOTH: an `<a>` for a status that has
   rows, and a `<span>` for one with a count of zero — because following a zero would empty the
   catalogue, and that screen's empty state means *"the seeds have not run"*. ▶ Which is the
   distinction above doing real work rather than decorating: on one screen, in one row of chips,
   the hand cursor is the only thing separating a filter from a read-out.

   ⚠ **AND THE RESTYLE ABOVE MADE IT WORSE BEFORE IT MADE IT BETTER.** Until then the `<a>`
   chips carried the browser's default underline and the `<span>` chips did not — so the
   underline was, by accident, the ONLY thing telling a reader which half could be clicked.
   Removing it for looking like "a paragraph of links that lost its styling" was right, and
   it deleted that signal. ▶ So the interactive affordances move onto the anchor, where they
   describe something true: a hand cursor and a hover response mean this one does something.
   A `<span>` chip now reads as what it is — a label with a figure. */
a.vz-chip {
  cursor: pointer;
}

/* 🔴 THE COUNT IS THE DATA AND IT WAS THE SAME SIZE AS ITS LABEL. A chip reads
   "Active: 5" — the word is the question and the figure is the answer, so the figure
   takes the ink and a step up in size while the label stays muted and small. That is
   the same hierarchy `.vz-widget-figure` uses on the board, one control smaller.
   ⚠ `tabular-nums` so a column of chips does not jitter as counts change width. */
.vz-chip strong {
  color: var(--vz-ink);
  font-size: var(--vz-fs-md);
  font-weight: var(--vz-fw-bold);
  font-variant-numeric: tabular-nums;
}

a.vz-chip:hover {
  border-color: var(--vz-primary-strong);
  background: var(--vz-surface-2);
}

/* 🔴 SEVERITY, BORROWED FROM `.vz-badge` RATHER THAN INVENTED. That control already
   draws an 8px dot and already owns the four variant names, and the two sit on the
   same screens — `/admin/organizations` renders `.vz-badge--success` on every row of
   the list the chips above it filter. A second dot vocabulary would let one screen
   describe one state two ways.
   ⚠ THE DOT IS SUPPLEMENTARY, NEVER THE SIGNAL. Each chip carries its status as a
   word already; the colour only lets somebody find the red one without reading five
   labels. Nothing here is conveyed by colour alone. */
.vz-chip::before {
  content: '';
  width: 8px;
  height: 8px;
  border-radius: 50%;
  background: var(--vz-border-strong);
  flex: 0 0 auto;
  margin-right: var(--vz-space-1);
}

.vz-chip--success::before { background: var(--vz-success); }
.vz-chip--info::before    { background: var(--vz-info); }
.vz-chip--warning::before { background: var(--vz-warning); }
.vz-chip--danger::before  { background: var(--vz-danger); }

/* ⚠ THE TINT IS NOT THE ONLY SIGNAL. `dashboard.js` also sets `aria-pressed`, and
   the border darkens as well as the fill — a state carried by colour alone is
   invisible to anybody who cannot separate these two greens. */
.vz-chip.is-active {
  border-color: var(--vz-primary-strong);
  background: var(--vz-primary-tint);
  color: var(--vz-primary-strong);
  font-weight: var(--vz-fw-medium);
}

.vz-widget-list {
  margin: var(--vz-space-3) 0 0;
  padding: 0;
  list-style: none;
}

/* 🐛 ⚠ `flex-wrap` AND A REAL BASIS ON THE TEXT — THIS IS P5's RIBBON, REPRODUCED.
   Without them the row is icon + text + Add button on one line at every width, and
   the §E6 sweep measured the description column at **116px carrying 104 characters**
   at 320px — a ten-line ribbon two words wide, with `documentElement.scrollWidth -
   clientWidth` reading **0 at all 1,121 widths**, because nothing was overflowing.
   `.vz-facts` was `minmax(8rem, 14rem) 1fr` and did exactly the same thing to three
   sentences on P5's screen.

   The basis is what makes the wrap happen: below ~14rem of available text column the
   action drops to its own line and the sentence takes the full width of the card. */
.vz-widget-card {
  display: flex;
  flex-wrap: wrap;
  align-items: flex-start;
  gap: var(--vz-space-3);
  padding: var(--vz-space-3) 0;
  border-top: 1px solid var(--vz-border);
}

/* Keeps the control in the same column as every other row's while they share a line,
   and hard against the right when it has wrapped onto its own. */
.vz-widget-card > form,
.vz-widget-card-state { margin-inline-start: auto; }

.vz-widget-card:first-child { border-top: 0; }

/* See `.vz-widget-panel-filters[hidden]` — the same reason, one component along.
   A card is `display: flex`, so `el.hidden` needs this rule to mean anything and
   a filtered-out card would otherwise stay on screen AND in the tab order. */
.vz-widget-card[hidden] { display: none; }

/* ⚠ `min-width: 0` SO THE TEXT COLUMN CAN SHRINK. A flex item defaults to its
   content's width, and one long widget name would otherwise push the Add button
   out of a 26rem panel — the failure `.vz-board`'s own `min-width: 0` exists to
   stop, one component along. */
.vz-widget-card-text { flex: 1 1 14rem; min-width: 0; }

.vz-widget-card-name {
  margin: 0;
  font-weight: var(--vz-fw-medium);
  color: var(--vz-ink);
}

.vz-widget-card-note {
  margin: var(--vz-space-1) 0 0;
  font-size: var(--vz-fs-sm);
  color: var(--vz-muted);
}

/* 🔴 ⚠ THE ADD CONTROL IS A ROUND "+", NOT A WORDED GHOST BUTTON (owner,
   2026-09-07). It was `.vz-action--quiet` reading *"Add Collections due"* — a
   borderless control on a card that already prints the widget's name twice, so
   it repeated the name a third time and read as a text link rather than as the
   thing to press.

   ⚠ THE ACCESSIBLE NAME DID NOT CHANGE. The words moved into a `.vz-sr-only`
   span, so this is a visual change and not a semantic one — an icon button with
   no text has no accessible name at all, which is the failure this avoids.

   ⚠ FULL `--vz-tap-min`, AND `box-sizing: border-box` BECAUSE IT SIZES AND PADS
   (`testEveryRuleThatSizesAndPadsSaysWhichBoxItMeans`). `align-self: center`
   matches `.vz-widget-card-state`, so a row's control sits in the same place
   whether the widget is on the board or not and the list does not jitter. */
.vz-widget-add {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  box-sizing: border-box;
  flex: 0 0 auto;
  align-self: center;
  min-width: var(--vz-tap-min);
  min-height: var(--vz-tap-min);
  padding: 0;
  border: 1px solid var(--vz-primary-strong);
  border-radius: 50%;
  background: var(--vz-primary-tint);
  color: var(--vz-primary-strong);
  cursor: pointer;
}

.vz-widget-add:hover {
  /* ⚠ `--vz-primary-button` AND `--vz-primary-contrast`, NEVER `--vz-primary`.
     E12 moved every white-on-green pair off that token because it reads 2.59 —
     below AA — and this is the same white label on the same green. */
  border-color: var(--vz-primary-button);
  background: var(--vz-primary-button);
  color: var(--vz-primary-contrast);
}

.vz-widget-add:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}

/* ⚠ A STATE, DRAWN AS TEXT AND NOT AS A BUTTON. It sits where the Add control
   sits so the row keeps its shape, and it must not look pressable — a disabled
   button there would invite a click that reports nothing. */
.vz-widget-card-state {
  flex: 0 0 auto;
  align-self: center;
  padding: var(--vz-space-2) var(--vz-space-3);
  font-size: var(--vz-fs-sm);
  font-weight: var(--vz-fw-medium);
  color: var(--vz-primary-strong);
}

/* The glyph's disc, tinted by the widget's CATEGORY rather than by its module —
   so a reader scanning the panel sees the filters they can press, drawn on every
   row. `currentColor` does the rest: `partials/icon` inherits it, which is why
   each rule below sets a colour and not a fill.

   ⚠ THE BOARD'S TILES DELIBERATELY DO NOT SHARE THIS TINT (§D12b). Extracting the
   five pairs into one set was the right instinct and the wrong fact: a KPI tile has
   **no room for a tinted box**, measured — see the head's own comment above — so a
   shared set would have had exactly one caller and a second name to keep in step.
   The gallery is a full-width row; that is why the box fits here and nowhere else. */
.vz-widget-card-icon {
  display: inline-flex;
  align-items: center;
  justify-content: center;
  flex: 0 0 auto;
  width: 2.25rem;
  height: 2.25rem;
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface-2);
  color: var(--vz-muted);
}

.vz-widget-card-icon--collections { background: var(--vz-info-tint);    color: var(--vz-info-text); }
.vz-widget-card-icon--chits       { background: var(--vz-primary-tint); color: var(--vz-primary-strong); }
.vz-widget-card-icon--members     { background: var(--vz-surface-2);    color: var(--vz-muted); }
.vz-widget-card-icon--money       { background: var(--vz-warning-tint); color: var(--vz-warning-text); }
.vz-widget-card-icon--compliance  { background: var(--vz-danger-tint);  color: var(--vz-danger-text); }

/* The board's own widget heading carries its glyph too, at text size and in the
   heading's own colour — it is a label, not a badge, and `§D12b` measured why it
   has to stay one. */
.vz-widget-title .vz-icon {
  flex: 0 0 auto;
  color: var(--vz-subtle);
}

/* 🐛 ⚠ FOUR ACROSS NEEDS A WIDE WINDOW, AND BELOW ONE A TILE IS TOO NARROW FOR
   ITS OWN CONTENTS. Measured in the §E6 sweep: at 768px and 1024px a quarter
   resolves to **126–128px** once the sidebar has taken its share, and
   `₹25,00,000.00` at `--vz-fs-display` is **clipped** there — `scrollWidth`
   exceeds `clientWidth` on the figure while the page reports **0px overflow at
   all 1,121 widths**, because nothing is overflowing: the number is simply cut
   off inside a box that fits. Row 14's rule, one component along — measure the
   inner box against its container, not only the document against the viewport.

   ⚠ TWO ACROSS RATHER THAN A SMALLER NUMBER. Shrinking the font would keep four
   tiles that no longer read as headlines; a tile's whole job is to be glanced at.
   At 1200px and up a quarter is ~186px, which fits the longest money figure this
   product formats. */
@media (max-width: 75rem) {
  .vz-widget--quarter { grid-column: span 6; }
}

@media (max-width: 44rem) {
  /* Every widget full width. ⚠ THE CATALOGUE'S WIDTH IS NOT REWRITTEN — a phone
     is a viewport, not a preference, and the same widget on a desktop still sits
     at the width its entry declares.

     ⚠ AND THIS BLOCK MUST STAY BELOW THE 75rem ONE. Both match on a phone, and
     the later rule wins at equal specificity — reversing them would put a tile
     back to half width at 375px. */
  .vz-widget--full,
  .vz-widget--half,
  .vz-widget--third,
  .vz-widget--quarter { grid-column: span 12; }
}

/* ===========================================================================
   Your own record, and the manual (chassis #17/#18)
   =========================================================================== */

/* ⚠ THE DISC IS THE TOPBAR'S OWN COMPONENT AT A LARGER SIZE, NOT A SECOND ONE.
   `.vz-avatar` is 32px because it sits in a menu summary; here it is the subject
   of a panel and needs to be looked at. Scaling the shared component keeps the
   contrast pair `ThemeContrastTest` already measures — a second disc would be a
   second pair nobody has proved (`§E12`). */
.vz-profile-avatar {
  display: flex;
  flex-wrap: wrap;
  align-items: flex-start;
  gap: var(--vz-space-5);
}

.vz-profile-avatar-disc .vz-avatar {
  width: 96px;
  height: 96px;
  font-size: var(--vz-fs-display);
}

.vz-profile-avatar-disc .vz-avatar--anon svg { width: 48px; height: 48px; }

/* The member's page: photo or initials beside the name. */
.vz-member-head {
  display: flex;
  align-items: center;
  gap: var(--vz-space-4);
  min-width: 0;
}

.vz-member-head-text { min-width: 0; }

.vz-member-photo-link { flex: none; display: block; border-radius: 50%; }

.vz-member-photo {
  position: relative;
  flex: none;
  display: block;
  width: 72px;
  height: 72px;
  border-radius: 50%;
  overflow: hidden;
}


.vz-member-photo .vz-avatar {
  width: 100%;
  height: 100%;
  font-size: var(--vz-fs-lg);
}

.vz-member-photo-image {
  position: absolute;
  inset: 0;
  width: 100%;
  height: 100%;
  object-fit: cover;
}

/* Grows into the room and CAN shrink, so a long refusal wraps instead of
   pushing the disc off the panel. */
.vz-profile-avatar-controls {
  flex: 1 1 16rem;
  min-width: 0;
  display: flex;
  flex-direction: column;
  gap: var(--vz-space-3);
  align-items: flex-start;
}

/* --- The manual ----------------------------------------------------------- */

/* ⚠ A LIST OF DESTINATIONS, NOT A TABLE. Each entry is a link and a sentence;
   below 44rem a `.vz-table` would become the members card grid, and a table of
   contents is not a data table in the first place. */
.vz-help-list { margin: 0; padding: 0; list-style: none; }

.vz-help-list li {
  padding: var(--vz-space-3) 0;
  border-top: 1px solid var(--vz-border);
}

.vz-help-list li:first-child { border-top: 0; }

.vz-help-link {
  color: var(--vz-primary-strong);
  font-weight: var(--vz-fw-medium);
  text-decoration: none;
}

.vz-help-link:hover { text-decoration: underline; }

.vz-help-tagline {
  margin: var(--vz-space-1) 0 0;
  font-size: var(--vz-fs-sm);
  color: var(--vz-muted);
}

/* 🔴 THE LINE A SEARCH MATCHED (F29) — the one thing the index cannot do.
   It sits UNDER the tagline and has to read as a QUOTATION rather than as a
   second tagline, or the result becomes three grey lines a reader has to parse.
   A left rule and the body ink do that without a new colour; the size stays
   `--vz-fs-sm` so the pair above it keeps the emphasis.
   ⚠ It is never rendered for a title or tagline hit — see `HelpSearch` — so this
   rule is only ever seen where it is carrying information. */
.vz-help-excerpt {
  margin: var(--vz-space-2) 0 0;
  padding-left: var(--vz-space-3);
  border-left: 2px solid var(--vz-border);
  font-size: var(--vz-fs-sm);
  color: var(--vz-ink);
}

/* The "other guides" list at the foot of a guide: names only, so it stays a
   way out rather than a second table of contents. */
.vz-help-list--compact {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2) var(--vz-space-4);
}

.vz-help-list--compact li { padding: 0; border-top: 0; }

.vz-help-intro { margin: 0 0 var(--vz-space-4); }

/* ⚠ THE NUMBERS ARE THE CONTENT, so they are rendered rather than suppressed —
   a click path whose order is invisible is a list of unrelated instructions. */
.vz-help-steps {
  margin: 0;
  padding-left: var(--vz-space-5);
}

.vz-help-steps li { margin-bottom: var(--vz-space-2); }
.vz-help-steps li:last-child { margin-bottom: 0; }

.vz-help-notes {
  margin: 0;
  padding-left: var(--vz-space-5);
}

.vz-help-notes li { margin-bottom: var(--vz-space-2); }
.vz-help-notes li:last-child { margin-bottom: 0; }

/* ===========================================================================
   The control plane's customers, databases, schema and queue (Q3)
   ---------------------------------------------------------------------------
   ⚠ FOUR CELL CLASSES AND FIVE TABLE MODIFIERS, AND BOTH NUMBERS ARE AS SMALL
   AS THE SCREENS ALLOW. Owner, 2026-09-08: `/admin` reuses the operator app's
   design system and must not start a second component set. So the operator's
   name reuses `.vz-c-name`, the badges reuse `.vz-badge`, the timestamps reuse
   `.vz-c-when` and the row actions reuse `.vz-c-actions` — every one of them
   doing the job it already does elsewhere.

   ⚠ WHAT IS NEW IS NEW BECAUSE NOTHING EXISTING SAYS WHAT IT IS. Reaching for
   `.vz-c-code` to hold a schema version, or `.vz-c-standing` to hold a
   provisioning step, would be `OPERATING_MODE §7` item 54 exactly: the name
   that is in the list rather than the name that says what the thing is.
   `§Q2` made the same argument about `.vz-c-hits` and named the three classes
   it was preferred over; these do the same below.

   🔴 AND A `grid-template-areas` BLOCK IS REQUIRED RATHER THAN OPTIONAL.
   Below 44rem every `.vz-table` row becomes a card laid out by named areas, and
   `§E6d` measured what one unnamed cell costs: the roster named `standing`,
   `.vz-c-standing` had no `grid-area`, the cell was auto-placed, and it pushed
   the page 32px at 320 while every other measurement read clean.
   `StylesheetStaticTest::testEveryTableModifierRenderedByAViewHasACssRule` is
   what makes that a rule rather than a habit.
   =========================================================================== */

/* The schema version — `v88`, and on the drift table the number a whole fleet is
   compared against.

   ⚠ `.vz-c-code` WAS THE OBVIOUS REACH AND IS WRONG: that class is `--vz-muted`,
   because a member code is an identifier somebody dictates rather than a figure
   they judge. A version is read to decide whether a tenant is behind, so it takes
   `--vz-ink` like the count in `.vz-c-hits`. Tabular figures for that class's
   reason too — v9 and v88 must not shift the column when the list is sorted. */
.vz-c-version {
  width: 1%;
  white-space: nowrap;
  font-variant-numeric: tabular-nums lining-nums;
  color: var(--vz-ink);
}

/* The outcome of the last connection check, and on the run log what a migration
   did to one tenant.

   ⚠ IT IS NOT `.vz-c-status`, AND THE REASON IS THAT BOTH APPEAR IN ONE ROW. The
   operators register carries the operator's STATUS and the database's last CHECK
   side by side; they are two different questions with two different answers, and
   sharing one class would give the two cells one `grid-area` and stack them on a
   phone. */
.vz-c-result { width: 1%; white-space: nowrap; }

/* The database a tenant lives in, with its provider and host beneath.

   ⚠ IT WRAPS WHERE `.vz-c-code` WOULD NOT. A schema name plus an RDS endpoint —
   `vazhai-prod-2.cluster-abc123.ap-south-1.rds.amazonaws.com` — is one unbreakable
   token in a `nowrap` cell, and that is exactly the shape `.vz-table--messaging`
   found scrolling its wrapper sideways with the action off-screen while page
   overflow measured zero. `overflow-wrap: anywhere` and not `word-break: break-all`
   for C3b's reason: it breaks a word only when there is no other opportunity. */
.vz-c-store {
  white-space: normal;
  overflow-wrap: anywhere;
}

/* The narrative cell — a provisioning step with its error beneath, or a version
   pair with the migration that stopped the run. It is this pair of tables' widest
   column and the one that takes the slack, which is `.vz-table--errors`'s finding
   about an exception message applied to two smaller sentences. */
.vz-c-detail {
  white-space: normal;
  overflow-wrap: anywhere;
}

/* 🐛 🔴 A MACHINE-WRITTEN STRING IN A PANEL BREAKS, AND THIS RULE EXISTS BECAUSE
   ONE PUSHED A WHOLE PAGE 145px SIDEWAYS.

   `migration_ledger.last_error` on the operator's detail page reads, in full:
   *"Migration [0088_backfill_branch_accounts_created_since_0036.sql] failed in set
   [org] on [org:3]: SQLSTATE[42S02]…"* — and that filename is a **52-character token
   with no break opportunity in it**. Measured on the `§E6` capture: the schema panel
   overflowed by 162px and the DOCUMENT by 145px at the narrow end, on a page with no
   table on it at all. Every test was green.

   ⚠ IT IS THE PAIRING THAT IS SCOPED, NOT `.vz-mono` ITSELF. That class is font and
   letter-spacing and nothing else, and it is worn by member codes, phone numbers,
   references and ports all over the product — giving all of them `anywhere` would let
   a ten-digit Operator ID break across two lines, which is worse than the scroll it
   would prevent. A `.vz-panel-note` that is ALSO `.vz-mono` is, in every use this
   product has, a sentence a machine wrote about a failure.

   ⚠ `anywhere` AND NOT `word-break: break-all` — C3b's rule, which `.vz-c-ref` carries:
   it breaks a word only when there is no other opportunity, so the rest of the sentence
   still wraps between its words. */
.vz-panel-note.vz-mono {
  overflow-wrap: anywhere;
}

/* 🐛 🔴 THE OPERATOR'S CELL BREAKS A LONG TOKEN, AND THAT IS A MEASURED FIX
   RATHER THAN A PREFERENCE. A database name — `vzc_wire_absent_database`,
   `vazhai-prod-2.cluster-abc123.ap-south-1.rds.amazonaws.com` — is ONE
   unbreakable token, and on its own it set this column's min-content at 208px
   and the whole table at 675px inside a 478px wrapper at 775px, with the action
   off the right edge and page overflow reading zero. Allowing it to break took
   the table to 576 before a single column was removed.

   ⚠ IT IS SCOPED TO THE ADMIN TABLES AND MUST NOT BECOME `.vz-c-name`'s BASE
   RULE. One portal along, that class holds a MEMBER'S NAME, and
   `overflow-wrap: anywhere` would break a person's name mid-word on a screen a
   collector reads at somebody's door — worse than the scroll it would fix.

   ⚠ `anywhere` AND NOT `word-break: break-all` — C3b's reason, which `.vz-c-ref`
   already carries: it breaks a word only when there is no other opportunity, so
   an operator's name still wraps between its words. */
.vz-table--operators .vz-c-name,
.vz-table--estate .vz-c-name,
.vz-table--drift .vz-c-name,
.vz-table--runlog .vz-c-name,
.vz-table--runs-admin .vz-c-name {
  overflow-wrap: anywhere;
}

/* 🐛🔴 F12 — ONE LONG ERROR MESSAGE STARVED THE JOB COLUMN (owner, 2026-09-23).
   `.vz-c-detail` carries the run's failure reason under its timestamp. Measured at
   1440px: the longest such cell held **503 characters** against a shortest of 20, and
   under `table-layout: auto` that single outlier row sized the column for all 25 —
   **Started took 676px of 958 (71%)** and Job was left 71px. `overflow-wrap: anywhere`
   above then turned the squeeze into mid-word shattering rather than a visible overflow:
   `provisioning_worker` needs **120px** and was given **41px** on the failures table and
   **54px** on the history one.

   ⚠ REPORTED AS ONE TABLE AND MEASURED AS TWO. Both carry `.vz-table--runlog`, so this
   rule reaches both — but a fix verified against the history table alone would have left
   the worse one broken.

   ▶ `table-layout: fixed` IS THE FIX BECAUSE IT TAKES THE DECISION AWAY FROM THE CONTENT.
   The widths below are declared, so a 503-character cell wraps inside its column instead
   of resizing it, and no row can starve another. 14rem gives the Job column 224px against
   the 120px its longest name needs — room for a longer one without returning here.

   ⛔ AND IT IS SCOPED TO THE WIDE BAND ON PURPOSE. Below 64rem every row becomes a card
   (`tr { display: grid }`), where `table-layout` means nothing and the defect does not
   exist — verified at 857px before and after. A rule applied at every width would be
   inert at the narrow ones and would invite somebody to "simplify" it later.

   ⏸ THE OTHER FIX CONSIDERED AND NOT TAKEN: moving the failure message out of the
   Started cell into its own full-width row. Better information architecture — a reason
   is not a timestamp — but a second `<tr>` becomes a SEPARATE CARD in card mode, so an
   error would detach from the run it belongs to at exactly the width where the screen is
   hardest to read. That fix needs a card-mode story first.

   ⚠ THE WIDTHS SIT ON `thead th`, NOT ON THE `.vz-c-*` CELLS, AND THAT IS NOT A STYLE
   CHOICE. Under `table-layout: fixed` the browser reads its column widths from the FIRST
   ROW ONLY and ignores every later one. Widths were first written against `.vz-c-name` /
   `.vz-c-result` / `.vz-c-actions` — the classes the `<td>`s carry — and the tables came
   back **305/305/305 and 240/240/240/240**: fixed layout applied, the widths did not, and
   the columns split evenly. The header cells carry no classes, and two of the history
   table's four are built by the `$sortHeader()` helper, so position is what there is to
   aim at.
   ⚠ `:nth-child(4)` IS INERT ON THE THREE-COLUMN FAILURES TABLE by construction — it has
   no fourth header — which is why one rule serves both tables without a second modifier. */
@media (min-width: 64.0625rem) {
  .vz-table--runlog { table-layout: fixed; }

  .vz-table--runlog thead th:first-child  { width: 14rem; }
  .vz-table--runlog thead th:nth-child(2) { width: 11rem; }
  .vz-table--runlog thead th:nth-child(4) { width: 6rem; }
}

@media (max-width: 64rem) {
  /* The customer register. The card leads with WHO and their state, because that
     is the pair somebody scanning a fleet is matching against, and the schema
     version takes the line beneath.

     ⚠ THERE IS NO `result` AREA HERE ANY MORE — see the view: the connection
     column was what took this table 197px past its wrapper, and `ADMIN_PANEL §2`
     never asked for one on this strip. */
  .vz-table--operators tr {
    grid-template-areas:
      'name    status'
      'version version'
      'actions actions';
  }

  /* The database estate. The store takes a full line of its own — see `.vz-c-store`
     for why a host is the one string here that must be allowed to break. */
  .vz-table--estate tr {
    grid-template-areas:
      'name    result'
      'store   store'
      'when    when'
      'actions actions';
  }

  /* 🐛 🔴 THE DRIFT CARD IS ONE COLUMN, AND IT IS THE ONLY ONE OF THE FIVE THAT
     IS — WHICH IS A MEASURED FIX FOR A TRAP IN THE SHARED CARD GRID.

     `.vz-table tr` is `grid-template-columns: minmax(0, 1fr) auto`, so the SECOND
     column is sized by its content and the first may shrink to **nothing**. This
     row's status cell carries a migration failure — a full SQL sentence — and at
     320 it took the whole card: `grid-template-columns` computed to **`0px 242px`**,
     the operator's name and its version both had `clientWidth: 0`, and their text
     spilled straight across the badge beside it. Page overflow read **zero**, the
     wrapper read zero, and every test was green.

     ⚠ THE `overflow-wrap: anywhere` ABOVE IS WHAT LET IT REACH ZERO RATHER THAN
     MERELY GET NARROW, which is worth saying plainly: the fix that closed the
     desktop scroll is what made the phone collapse visible. Both are real and both
     are now measured.

     ⚠ AND `minmax(0, 1fr) auto` STARVING ITS FIRST COLUMN IS A PROPERTY OF THE
     SHARED RULE RATHER THAN OF THIS TABLE — it bites any card whose second column
     holds a sentence. Only this row's is fixed here; the general finding is written
     into `START_HERE §6` rather than corrected across thirteen registers inside a
     Phase Q row (`OPERATING_MODE §7` item 63's boundary, one defect along).

     ⚠ TWO CONTROLS SIT IN THE ACTION CELL, AND `.vz-c-actions` IS A WRAPPING FLEX
     ROW FROM B5, so they stack rather than pushing the card. */
  .vz-table--drift tr {
    grid-template-areas:
      'name    name'
      'version version'
      'status  status'
      'actions actions';
  }

  /* What the last run did, per tenant. No action cell — this table is a report of
     something that already happened. */
  /* 🐛 🔴 `actions` WAS ADDED IN Q4, AND ITS ABSENCE WAS `§E6d`'s DEFECT EXACTLY.
     Q3 wrote this modifier for a run-log table that has no row actions, and Q4
     reused it for the worker's history — which HAS one. `.vz-c-actions` carries
     `grid-area: actions` globally, so a template that names no such area put that
     cell in an IMPLICIT column: measured at 320, the row computed to
     `60.58px 84.88px 0px 72.55px` — **four tracks where the template names two**,
     with the action auto-placed in one nobody chose.

     ⚠ AND EVERY OTHER MEASUREMENT READ CLEAN — page overflow 0, the wrapper 0,
     every test green — because the implicit column happened to fit. `§E6d` found
     the same shape by the sideways push it caused; this one was found by reading
     `gridTemplateColumns` BACK, which is the check `START_HERE §6` finding 8 asks
     for and the only thing that could have seen it.

     ⚠ AN AREA WITH NO CELL IN IT COLLAPSES TO NOTHING, so the three tables that
     use this modifier WITHOUT actions — the migrations run log, the failures list
     and the abandoned-runs panel — are unaffected. That is what makes one modifier
     for four tables cheaper than four modifiers to keep in step. */
  .vz-table--runlog tr {
    grid-template-areas:
      'name    result'
      'detail  detail'
      'actions actions';
  }

  /* The provisioning queue, and the stranded panel above it.

     ⚠ ONE MODIFIER SERVES BOTH TABLES ON THAT PAGE, DELIBERATELY. The stranded
     list renders no `detail` cell, and a named area with nothing in it simply
     collapses — where a second modifier would be a second place for the same page's
     two tables to drift apart. */
  .vz-table--runs-admin tr {
    grid-template-areas:
      'name   status'
      'detail detail'
      'when   when';
  }

  /* ⚠ `width: auto` UNDOES THE SHRINK-TO-FIT ON BOTH NARROW CELLS: these are grid
     areas now, not table cells, and 1% of a card is 3px. `.vz-c-actions` carries
     the same correction and records the measurement. */
  .vz-c-version { grid-area: version; width: auto; }
  .vz-c-result  { grid-area: result; width: auto; justify-self: end; text-align: right; }

  /* ⚠ THE DRIFT CARD'S CELLS ARE ALL FULL-WIDTH, so nothing is pushed to the end —
     `justify-self: end` on a single-column row would right-align a sentence. */
  .vz-table--drift .vz-c-status { justify-self: start; text-align: left; }
  .vz-c-store   { grid-area: store; }
  .vz-c-detail  { grid-area: detail; }

  /* The operator's name leads every one of these cards, so it takes the card's
     heading size — `.vz-table--errors`'s treatment of its problem cell. */
  .vz-table--operators .vz-c-name,
  .vz-table--estate .vz-c-name,
  .vz-table--drift .vz-c-name,
  .vz-table--runlog .vz-c-name,
  .vz-table--runs-admin .vz-c-name {
    font-size: var(--vz-fs-md);
    white-space: normal;
  }

  /* ⚠ THE TIMESTAMP WRAPS IN THESE TABLES, WHICH IS `.vz-table--errors`'s MEASURED
     FIX AND NOT A PREFERENCE. `.vz-c-when`'s base rule is `nowrap`, right on nine
     registers where the date sits beside short cells; here "8 Sep 2026, 11:46 am"
     is 162px of unbreakable text in a card two columns wide. */
  .vz-table--estate .vz-c-when,
  .vz-table--drift .vz-c-when,
  .vz-table--runs-admin .vz-c-when {
    white-space: normal;
  }
}

/* The sub-lines take the whole width in all five, so the operator id, the host and
   the error each sit under the thing they qualify rather than beside it.
   `.vz-sub-meta` already draws that relationship everywhere else in the product;
   this only tells it to take the line. */
.vz-table--operators .vz-c-name .vz-sub-meta,
.vz-table--estate .vz-c-name .vz-sub-meta,
.vz-table--estate .vz-c-store .vz-sub-meta,
.vz-table--drift .vz-c-name .vz-sub-meta,
.vz-table--drift .vz-c-status .vz-sub-meta,
.vz-table--drift .vz-c-version .vz-sub-meta,
.vz-table--runlog .vz-c-detail .vz-sub-meta,
.vz-table--runs-admin .vz-c-name .vz-sub-meta,
.vz-table--runs-admin .vz-c-status .vz-sub-meta,
.vz-table--runs-admin .vz-c-detail .vz-sub-meta,
.vz-table--runs-admin .vz-c-when .vz-sub-meta {
  display: block;
}

/* ⚠ THE DRIFT TABLE'S STATUS CELL CARRIES AN ERROR MESSAGE UNDER ITS BADGE, so it
   is the one `.vz-c-status` in the product that must be allowed to wrap — the base
   rule keeps a badge on one line, which is right everywhere a status is only a
   status. */
.vz-table--drift .vz-c-status,
.vz-table--runs-admin .vz-c-status {
  white-space: normal;
}

/* The version cell on the drift table holds "v88" and "Pending: 1" as two lines, so
   the shrink-to-fit width has to give. */
.vz-table--drift .vz-c-version,
.vz-table--operators .vz-c-version {
  white-space: normal;
}

/* ===========================================================================
   The control plane's background worker (Q4)
   ---------------------------------------------------------------------------
   ⚠ NO NEW CELL CLASSES AT ALL, WHICH IS WHAT Q3's FOUR BOUGHT. The worker's
   three tables are built from `.vz-c-name`, `.vz-c-result`, `.vz-c-detail`,
   `.vz-c-when` and `.vz-c-actions` — every one of them already doing the job it
   does elsewhere. `.vz-table--runlog` is reused unchanged for the history, the
   failures list and the abandoned-runs panel: three tables, one shape, one card
   layout to keep in step.

   ⚠ TWO NEW MODIFIERS, AND BOTH EXIST ONLY FOR THEIR `grid-template-areas`.
   Below 44rem every `.vz-table` row becomes a card laid out by named areas, and
   `§E6d` measured what one unnamed cell costs. `StylesheetStaticTest::
   testEveryTableModifierRenderedByAViewHasACssRule` is what makes that a rule.
   =========================================================================== */

@media (max-width: 64rem) {
  /* The locks panel. The holder is `host:pid:nonce` — the string somebody copies
     into an ssh command — so it takes a full line rather than sharing one with a
     timestamp. */
  .vz-table--locks tr {
    grid-template-areas:
      'name   name'
      'detail detail'
      'when   when';
  }

  /* The scheduled set. Two cells, and the second carries a badge, a moment and a
     count triple — so it is a line of its own rather than a column beside a job
     name that is itself two lines. */
  .vz-table--schedule-admin tr {
    grid-template-areas:
      'name   name'
      'detail detail';
  }
}

/* 🔴 THE WORKER'S TABLES BREAK A LONG TOKEN, FOR Q3's MEASURED REASON. A lock
   holder is `some-host-name:31337:a1b2c3`, a job's `last_error` is
   `App\Services\Whatever: SQLSTATE[…]`, and both are single unbreakable tokens
   that set a column's min-content on their own — which is what took Q3's
   operators register 675px into a 478px wrapper before `anywhere` was applied.

   ⚠ SCOPED TO THESE TABLES, never to `.vz-c-name` itself: one portal along that
   class holds a MEMBER'S NAME, and breaking a person's name mid-word would be
   worse than the scroll it fixed. */
.vz-table--locks .vz-c-name,
.vz-table--locks .vz-c-detail,
.vz-table--schedule-admin .vz-c-name,
.vz-table--schedule-admin .vz-c-detail {
  overflow-wrap: anywhere;
}

/* The sub-lines take the whole width, so a job's description sits under its name
   and a lock's "taken at" under its holder rather than beside them. `.vz-sub-meta`
   already draws that relationship everywhere else in the product. */
.vz-table--locks .vz-c-name .vz-sub-meta,
.vz-table--locks .vz-c-detail .vz-sub-meta,
.vz-table--schedule-admin .vz-c-name .vz-sub-meta,
.vz-table--schedule-admin .vz-c-detail .vz-sub-meta {
  display: block;
}

/* The lock name and the scheduled job name are read character by character, so
   the cell must be allowed to wrap where `.vz-c-name`'s base rule would not
   force it to — and the badge beneath needs the line. */
.vz-table--locks .vz-c-name,
.vz-table--schedule-admin .vz-c-name {
  white-space: normal;
}

/* ⚠ THE DISCLOSURE'S NESTED LISTS SIT INSIDE A `.vz-kv` VALUE, and a browser's
   default `<ul>` padding would indent them off a 320px card. `.vz-kv` is a
   description grid, so the list is a block inside `dd` rather than a sibling of
   it — this only removes the indent and the bullets, which carry no meaning when
   every item is one `key: value` pair. */
.vz-kv dd > ul {
  margin: 0;
  padding: 0;
  list-style: none;
}

.vz-kv dd > ul > li + li { margin-top: var(--vz-space-2); }

/* ===========================================================================
   Messaging — the template catalogue, the queue panel and the fleet outbox (Q5)
   ---------------------------------------------------------------------------
   🐛 🔴 THREE NEW MODIFIERS, AND THE DRAFT HAD ONE — the sweep is what added
   the other two.

   The outbox draft wore `.vz-table--messaging`, the operator message queue's,
   reasoning that one modifier for two tables answering the same question is
   cheaper than two kept in step (`§Q4`'s argument for `--runlog`). MEASURED AND
   REFUSED: row 14 pinned `.vz-c-name` to `width: 18ch; min-width: 14ch` because
   "an email address is one unbreakable word — branch.manager@coimbatore.example
   .test measures 312px" and `overflow-wrap: anywhere` had crushed it to ~1ch.
   THIS column holds a PHONE NUMBER, whose min-content is 103px and which needs
   no such floor — and no class distinguishes the two tables, so the floor could
   not be overridden without inventing one.

   ⚠ A REUSE IS ONLY FREE WHEN THE CONTENT IS THE SAME SHAPE, WHICH IS NARROWER
   THAN "THE SAME QUESTION". The version-history table DOES reuse
   `.vz-table--runs-admin`, because three cells of the same kinds fit its areas
   exactly and it renders no cell that modifier names no area for — which is the
   safe direction of reuse and the one `§Q4` got wrong.
   =========================================================================== */

/* 🔴 THE CATALOGUE'S LOCALE CELL IS ONE COLUMN, NOT ONE COLUMN PER LOCALE, AND
   THAT IS A LAYOUT DECISION MADE FROM `§Q2`'s MEASUREMENT RATHER THAN FROM
   TASTE.

   `ADMIN_PANEL §2` asks for "a per-locale view", and the obvious rendering is a
   column per language. Eight locales are seeded, which is ten columns — and
   `§Q2` measured what SEVEN costs: 737px of table inside a 553px wrapper,
   scrolling sideways from 705px to about 1030px with the action off the right
   edge, while page overflow read ZERO at every one of those widths.

   ⚠ AND A COLUMN COUNT THAT COMES OUT OF THE DATABASE CANNOT HAVE A
   `grid-template-areas` BLOCK AT ALL. Below 44rem every `.vz-table` row becomes
   a card laid out by named areas; a cell with no area is auto-placed into an
   implicit column, which is `§E6d`'s defect and Q4's. There is no way to name an
   unknown number of areas, so the card layout would have been broken by
   construction for every language after the first.

   So the locales share one cell, one line each: two fixed columns whatever
   `languages` holds, a writable card layout, and a new language grows the cell
   DOWNWARDS rather than the table sideways. */
.vz-c-locales {
  white-space: normal;
}

/* The locale lines stack, each one reading "code — state". `.vz-sub-meta` draws
   that relationship everywhere else in the product; this only tells it to take
   the whole line so a badge and its language code stay together. */
.vz-c-locales .vz-sub-meta {
  display: block;
}

.vz-c-locales .vz-sub-meta + .vz-sub-meta {
  margin-top: var(--vz-space-1);
}

/* 🐛 THE CATEGORY WAS A COLUMN AND IS NOW A SUB-LINE, MEASURED: three columns
   put this table at 569px inside a 436px wrapper at 775 — 133px of sideways
   scroll with page overflow reading -15, because `.vz-table-wrap` absorbs it.
   `.vz-c-kind` is `white-space: nowrap` and "Authentication" is one word, so
   relaxing the width moved NOTHING. `§Q2`'s rule again: the answer is one fewer
   column. It belongs on the template's own line anyway, beside the channel.

   A template code is `auction_rescheduled` — one unbreakable 19-character token
   — so the cell wraps where `.vz-c-name`'s base rule would hold it on one line.
   Scoped to this table: one portal along that class holds a MEMBER'S NAME, and
   breaking a person's name mid-word would be worse than the scroll it fixed. */
.vz-table--templates .vz-c-name {
  white-space: normal;
}

.vz-table--templates .vz-c-name .vz-sub-meta {
  display: block;
}

/* The waiting panel's template key is the whole `code · locale · vN` tuple —
   what `OutboxSendJob` resolves exactly, with no fallback — and it is the one
   string on the panel somebody reads character by character. */
.vz-table--waiting .vz-c-name {
  white-space: normal;
  overflow-wrap: anywhere;
}

.vz-table--waiting .vz-c-name .vz-sub-meta,
.vz-table--waiting .vz-c-status .vz-sub-meta {
  display: block;
}

/* 🐛 🔴 THE FLEET OUTBOX: FOUR COLUMNS, AND FIVE WERE MEASURED AND REFUSED.
   At a 775px viewport the content column is 478px and the sidebar has appeared:

     six columns (the operator in one of its own):        797px  (+319)
     five, every width relaxed and the date wrapping:     562px  (+84)
     four:                                                478px  (0px)

   Page overflow read -15 THROUGHOUT. `.vz-table-wrap` is `overflow-x: auto`, so
   it absorbs the scroll and the page sweep is right and blind — `§Q2`'s finding,
   and the reason `START_HERE §6` asks for the wrapper to be swept separately.

   ⚠ NO `min-width` FLOOR ON THE RECIPIENT BEYOND ITS OWN CONTENT. Row 14 needed
   `14ch` because an email address crushed to 1ch renders as 38 stacked letters;
   a phone number's min-content is 103px on its own and `+919876500011` must not
   break mid-number — a number split across two lines is a number somebody
   copies wrongly, which is `.vz-audit-values`' argument about a JSON value. */
.vz-table--outbox .vz-c-kind {
  width: auto;
  white-space: normal;
  overflow-wrap: anywhere;
}

.vz-table--outbox .vz-c-name {
  width: auto;
  /* The phone number itself, so the operator's name below may break and the
     number may not. */
  min-width: 11ch;
  white-space: normal;
  overflow-wrap: anywhere;
}

/* 🔴 THE STATE CELL CARRIES THE SENDER'S OWN SENTENCE, so it is the column that
   takes the slack — `.vz-table--messaging`'s finding and `.vz-table--errors`'.
   `SCHEMA_MASTER §3b`: `last_error` on a `queued` row is the HOLD reason, "and
   that is where 'Waiting' stops being a blank fact", so the badge and the
   explanation of the badge are one cell rather than two.

   ⚠ `white-space: normal` UNDOES `.vz-c-status`'s BASE `nowrap`, which is right
   on thirty registers where that cell holds one badge and wrong on the one where
   it holds a paragraph. The floor keeps it readable when the table has to fit. */
.vz-table--outbox .vz-c-name .vz-mono { overflow-wrap: anywhere; }

.vz-table--outbox .vz-c-status {
  min-width: 14ch;
  white-space: normal;
  overflow-wrap: anywhere;
}

.vz-table--outbox .vz-c-kind .vz-sub-meta,
.vz-table--outbox .vz-c-name .vz-sub-meta,
.vz-table--outbox .vz-c-status .vz-sub-meta {
  display: block;
}

/* ⚠ THE ATTEMPT COUNT WRAPS HERE. `.vz-c-moved`'s base rule is `nowrap`, right
   for the short "moved to 3 Sep" it was written for and wrong for "not yet
   attempted" — which was 110px of unbreakable run and the last 12px of the
   overflow above. */
.vz-table--outbox .vz-c-moved {
  white-space: normal;
}

/* 🐛 THE RAISED INSTANT WRAPS, WHICH IS `.vz-table--errors`'s MEASURED FIX.
   `.vz-c-when`'s base rule is `white-space: nowrap`, and "9 Sep 2026, 09:00 am"
   is 89px of unbreakable date beside three cells that are all at min-content. */
.vz-table--outbox .vz-c-when {
  white-space: normal;
}

@media (max-width: 64rem) {
  /* The catalogue card leads with WHICH TEMPLATE, because that is what somebody
     scanning the set is matching against — then the languages take a full line
     of their own, since that cell is a list rather than a value. */
  .vz-table--templates tr {
    grid-template-areas:
      'template template'
      'locales  locales';
  }

  .vz-table--templates .vz-c-name    { grid-area: template; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--templates .vz-c-locales { grid-area: locales; }

  /* The waiting card leads with the template key, then the count beside it,
     then the approval state on its own line — because the state is the only
     part with a link in it, and a control that navigates should not sit next to
     a number a thumb is scrolling past. */
  .vz-table--waiting tr {
    grid-template-areas:
      'template hits'
      'state    state';
  }

  /* ⚠ `width: auto` UNDOES THE SHRINK-TO-FIT ON THE COUNT: these are grid areas
     now, not table cells, and 1% of a card is 3px. `.vz-c-version` and
     `.vz-c-actions` both carry the same correction with the same measurement. */
  .vz-table--waiting .vz-c-name   { grid-area: template; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--waiting .vz-c-hits   { grid-area: hits; width: auto; justify-self: end; }
  .vz-table--waiting .vz-c-status { grid-area: state; justify-self: start; text-align: left; white-space: normal; }

  /* 🔴 🐛 THE OUTBOX CARD IS ONE COLUMN, AND `OPERATING_MODE §7` ITEM 64 IS WHY.
     `.vz-table tr` is `grid-template-columns: minmax(0, 1fr) auto`, so the
     SECOND column is sized by its content and the first may shrink to NOTHING.
     This row's state cell holds the sender's whole sentence — "Template
     [payment_reminder/whatsapp/en v1] is draft, not approved; awaiting Meta
     approval via Spring Edge." — which is exactly the shape that took
     `.vz-table--drift` to `grid-template-columns: 0px 242px` at 320, with two
     cells' text spilling across the badge beside them while page overflow, the
     wrapper AND every test read clean.

     ⚠ AND THE `overflow-wrap: anywhere` ABOVE IS WHAT WOULD LET IT REACH ZERO
     rather than merely get narrow — item 63's fix making item 64 reachable, one
     table along. A one-column card is the per-table answer `§6` finding 8 names,
     and `.vz-table--drift` is the precedent. */
  .vz-table--outbox tr {
    grid-template-areas:
      'message message'
      'to      to'
      'state   state';
  }

  .vz-table--outbox .vz-c-kind   { grid-area: message; width: auto; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--outbox .vz-c-name   { grid-area: to; width: auto; }
  .vz-table--outbox .vz-c-status { grid-area: state; justify-self: start; text-align: left; }

  /* ⚠ THE RAISED INSTANT IS DROPPED ON A PHONE, WITH AN EXPLICIT RULE AND NOT BY
     BEING LEFT UNPLACED — `.vz-table--messaging`'s answer to the same question,
     and its reason transfers unchanged: an unplaced cell still renders,
     auto-placed in a position nobody chose, and measures zero overflow because
     it fits (B5's blind spot). It is the one column a card can lose: the list is
     newest-first, so position already carries roughly when. */
  .vz-table--outbox .vz-c-when { display: none; }
}

/* 🔴 THE TEMPLATE'S WORDING, AND IT IS NOT `.vz-audit-values`.
   That class is this product's existing `<pre>` and it is `white-space: pre`
   for a stated reason — a JSON blob "has no spaces worth breaking at, so it
   scrolls rather than wrapping mid-token", because "a value split across two
   lines is a value somebody copies wrongly".

   A message body is the opposite kind of string: prose with spaces, which a
   person has to READ to review. Left on `pre` it would scroll sideways inside
   the panel and the reviewer would approve half a sentence.

   ⚠ SO `pre-wrap` PLUS `overflow-wrap: anywhere`, AND THE SECOND HALF IS NOT
   BELT AND BRACES: a WhatsApp body can carry a URL or a long token with no
   break opportunity, which is exactly the shape that pushed a whole page 145px
   sideways in Q3 with no table involved (a 52-character migration filename).
   `anywhere` rather than `word-break: break-all` — C3b's rule: it only breaks a
   word when there is no other opportunity, so ordinary prose still wraps at its
   spaces. */
.vz-wording {
  margin: 0 0 var(--vz-space-3);
  padding: var(--vz-space-3);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface-2);
  color: var(--vz-ink);
  font-family: var(--vz-font-mono);
  font-size: var(--vz-fs-sm);
  line-height: 1.6;
  white-space: pre-wrap;
  overflow-wrap: anywhere;
}

/* ===========================================================================
   Announcements — Vazhai → operator broadcasts (Q7)
   =========================================================================== */

/* 🔴 FIVE COLUMNS, AND THE TYPE AND SEVERITY ARE FOLDED INTO THE FIRST ONE
   BEFORE THE SWEEP RATHER THAN AFTER IT. `.vz-table--errors` learned that the
   hard way — seven columns was the defect and seven widths were the symptom,
   and folding the reference into the problem cell is what actually fixed it.
   The type and the severity are read WITH the title ("the maintenance one, the
   critical one") and are never scanned down a column of their own, so they ride
   inside `.vz-c-name` as `.vz-sub-meta` from the first draft.

   ⚠ EVERY CELL CLASS HERE ALREADY EXISTS GLOBALLY and was grepped before it was
   typed — `.vz-c-name`, `.vz-c-kind`, `.vz-c-when`, `.vz-c-status`,
   `.vz-c-actions`. `§6`'s house rule: an invented modifier leaves the base class
   drawing the box, so it measures as 0px and looks clean. */

/* 🐛 🔴 THE WINDOW WRAPS IN THIS TABLE, AND IT IS THE SAME MEASURED FIX
   `.vz-table--errors` needed for the same reason. `.vz-c-when`'s base rule is
   `white-space: nowrap`, which is right on the registers where the date sits
   beside short cells. Here the cell holds TWO instants — "15 Sep 2026, 2:00 am"
   over "until 15 Sep 2026, 5:00 am" — and unbreakable it is by some distance the
   widest column on the screen. ⚠ THE SECOND LINE IS ALREADY A BLOCK
   (`.vz-sub-meta` below), so this only relaxes the first. */
.vz-table--announcements .vz-c-when {
  white-space: normal;
}

/* The two ends of the window are one fact read together, so the second sits
   under the first rather than beside it. `.vz-sub-meta` already draws that
   relationship everywhere else in the product; this only tells it to take the
   whole line, exactly as `.vz-table--errors .vz-c-problem .vz-sub-meta` does. */
.vz-table--announcements .vz-c-when .vz-sub-meta,
.vz-table--announcements .vz-c-name .vz-sub-meta,
.vz-table--announcements .vz-c-kind .vz-sub-meta {
  display: block;
}

/* ⚠ THE TITLE IS THE COLUMN THAT TAKES THE SLACK. It is authored prose of up to
   200 characters and it is the whole reason somebody opened the screen, while
   every other cell is short and fixed: an audience word, two instants, a badge.

   ⚠ `overflow-wrap: anywhere` AND NOT `word-break: break-all` — C3b's rule,
   which `.vz-c-ref` and `.vz-table--errors .vz-c-problem` both carry: it breaks
   a word only when there is no other opportunity. A title is the one field in
   this module a person can paste a URL into. */
.vz-table--announcements .vz-c-name {
  white-space: normal;
  overflow-wrap: anywhere;
}

/* 🐛 🔴 THERE IS NO `.vz-c-kind` RULE HERE ANY MORE, AND ITS ABSENCE IS A
   MEASUREMENT. The draft gave the audience a column of its own and the §E6
   wrapper sweep refused it: **507px of table inside a 472px wrapper at a 769px
   viewport**, 35px of sideways scroll with the actions off-screen, and 29px still
   at 775 — the band §6 records as where §Q5's worst reading was. Page overflow
   read `-15` throughout, because `.vz-table-wrap` is `overflow-x: auto`.

   ⚠ AND RELAXING A WIDTH COULD NOT HAVE FIXED IT — the five columns measured
   107 · 95 · 90 · **127** · 87 and the widest was `.vz-c-status`, whose size is a
   `.vz-badge` reading "Archived" (90px of the design system's own padding) plus
   32px of cell padding. §6's rule for the fourth time: one fewer column, not one
   relaxed width. The audience now rides inside `.vz-c-name` with the type and the
   severity, where it is read anyway. */

@media (max-width: 64rem) {
  /* The card leads with the announcement, because that is what a person scans
     for — and the type, the severity and the audience all ride inside that cell
     rather than beside it. The window and the status then share a line, since
     "when does it show" and "has it gone out" are read together.

     ⚠ THREE ROWS RATHER THAN FOUR, BECAUSE THE SWEEP TOOK THE AUDIENCE COLUMN —
     see the note above the media query. An area named here with no cell to fill
     it collapses harmlessly (§6 finding 11), but leaving one behind would say
     this table still renders something it does not.

     🔴 ⚠ EVERY CELL THE VIEW RENDERS IS NAMED HERE. `§6` findings 8, 9 and 11:
     `.vz-c-actions` carries `grid-area: actions` GLOBALLY, so a modifier whose
     `grid-template-areas` did not name that area would put the cell in an
     IMPLICIT column — measured in Q4 as four tracks where the template named
     two, with page overflow, the wrapper and every test reading clean. */
  .vz-table--announcements tr {
    grid-template-areas:
      'name    name'
      'when    status'
      'actions actions';
  }

  .vz-table--announcements .vz-c-name    { grid-area: name; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--announcements .vz-c-status  { grid-area: status; justify-self: end; text-align: right; }
  .vz-table--announcements .vz-c-when    { grid-area: when; width: auto; justify-self: start; text-align: left; }
  .vz-table--announcements .vz-c-actions { grid-area: actions; text-align: left; }
}


/* ===========================================================================
   Reference data, locales and platform config (Q10)
   ---------------------------------------------------------------------------
   THREE TABLES, THREE MODIFIERS, AND NOT ONE NEW CELL CLASS. `.vz-c-name`,
   `.vz-c-code`, `.vz-c-status` and `.vz-c-detail` were all in this file before
   this row, and each was GREPPED before it was typed — `§6`'s note that this
   batch has added exactly two cell classes in nine rows, and Q7 and Q9 each
   added none.

   ⚠ EVERY ONE OF THE THREE STILL NEEDS ITS OWN MODIFIER, because the base
   `.vz-table tr` card grid is the MEMBERS grid and its areas are member column
   names. A new table inheriting them by accident is the trap that block's own
   comment warns about: the cells land in an order nobody chose and the layout
   still measures zero overflow, because it fits.
   =========================================================================== */

@media (max-width: 64rem) {
  /* The operator's own subscription invoices (`§Q-R4`, `§7` item 77) — the nineteenth
     module, and the only screen in `/app` whose rows belong to Vazhai.

     ⚠ THE SERIAL LEADS AND THE STATUS SITS OPPOSITE IT, because "VTC/2026-27/0007 —
     Paid" is the line somebody scans for. The period is the second row on its own:
     "1 Aug 2026 to 31 Aug 2026" is long and has nothing to share a row with.

     ⚠ AND THE TWO MONEY CELLS SHARE THE LAST ROW, total left and received right, so
     the comparison the card exists for is one glance rather than two. `.vz-c-money`
     already carries an unscoped `grid-area: value`, so it is named here explicitly —
     the `.vz-table--dues` block records what happens to a table that forgets to. */
  /* Payments against one of Vazhai's own invoices (`§Q-R4`, `§7` item 77).
     Four cells: the day, the amount, how it came in, and the bank reference.

     ⚠ THE AMOUNT SITS OPPOSITE THE DATE, because "₹4,720 on 15 Sep" is the one
     line somebody reconciling reads — and the reference takes a full row of its
     own, since a UTR is a long unbroken string with nothing to wrap on.

     ⚠ `.vz-c-money` ALREADY HAS AN UNSCOPED CARD RULE (`grid-area: value`), and
     the comment at the `.vz-table--dues` block records what happens when a table
     forgets to pair its own area with it: the cell falls through to `value`, an
     area this grid does not define, and lands wherever auto-placement puts it.
     **So `.vz-c-money` is named here explicitly rather than inherited.** */
  /* 🐛 PRICING PART 4 ADDED THREE CELLS — the receipt, the TDS and the receipt PDF — and the
     grid above named four, so at 375px the new ones were auto-placed on top of the old ones and
     the page ran 547px wide (found driving). ▶ Every cell has an area again: the receipt (with
     its Cancelled badge) under the date and amount it belongs to, the TDS with its section and
     certificate on its own row, and the PDF last. */
  .vz-table--payments tr {
    grid-template-areas:
      'when  amount'
      'code  code'
      'tds   tds'
      'mode  mode'
      'ref   ref'
      'doc   doc';
  }

  .vz-table--payments .vz-c-when   { grid-area: when; color: var(--vz-ink); }
  .vz-table--payments .vz-c-money  { grid-area: amount; justify-self: end; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--payments .vz-c-code   { grid-area: code; white-space: normal; }
  .vz-table--payments .vz-c-tds    { grid-area: tds; }
  .vz-table--payments .vz-c-document { grid-area: doc; justify-self: start; }
  .vz-table--payments .vz-c-document:empty { display: none; }
  .vz-table--payments .vz-c-mode   { grid-area: mode; }
  /* ⚠ `overflow-wrap: anywhere` AND A FLOOR WITH IT (row 14's finding). A UTR has
     no break opportunity, so without the wrap it pushes the card sideways — and
     with the wrap alone its min-content becomes ONE CHARACTER and it can be
     crushed into a vertical stack of letters. The pair is the fix, not either half. */
  .vz-table--payments .vz-c-ref {
    grid-area: ref;
    min-width: 8rem;
    overflow-wrap: anywhere;
    color: var(--vz-subtle);
    font-size: var(--vz-fs-xs);
  }

  /* The subdivisions: name, ISO code, active. Three cells, and the code sits
     under the name because they are read together — "Tamil Nadu, IN-TN" is one
     answer to one question, and putting the code opposite the badge would pair
     it with something it has nothing to do with. */
  .vz-table--states tr {
    grid-template-areas:
      'name status'
      'code code'
      'act  act';
  }

  .vz-table--states .vz-c-name   { grid-area: name; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--states .vz-c-code   { grid-area: code; }
  .vz-table--states .vz-c-status { grid-area: status; justify-self: end; }

  /* 🔴 THE RENAME CELL GETS ITS OWN ROW, ADDED WITH THE CONTROL (`§Q-R3`, 2026-09-15).
     A `td` with no area lands wherever the auto-placement puts it, at zero overflow,
     which is row 14's blind spot — so the areas grew in the same change as the cell.

     ⚠ IT HOLDS A LINK, NOT A FORM, AND THAT IS THE SECOND VERSION OF THIS RULE. The
     first draft put the whole rename form in this cell inside a `<details>`; measured
     with every `<details>` open, the table went from 916px inside a 916px wrap to
     1077 inside 901 and scrolled at every desktop width — `§Q-H7`'s "a cell holding a
     form floors its column", which no rule here could have fixed. The form is a routed
     page now and the cell is a 44px tap target.

     ⚠ `width: auto` AND `text-align: left` because `.vz-c-actions` is sized for icon
     buttons at the right edge of a wide table, and on a card it is a full-width row
     with a labelled control in it. */
  .vz-table--states .vz-c-actions {
    grid-area: act;
    width: auto;
    text-align: left;
  }

  /* The seed set. The catalogue cell carries three lines of its own — the count,
     what is missing, and the template count — so it takes a full row rather than
     sharing one with the badge.

     ⚠ `width: auto` AND `white-space: normal` ON THE DETAIL CELL for the reason
     `.vz-table--users .vz-c-role` carries: this is a grid area now rather than a
     table cell, and a `nowrap` list of five missing group names would push the
     card sideways at 320px. */
  /* ⚠ THE FOURTH AREA ARRIVED WITH THE WRITER (`§Q-R7`), AND AN UNPLACED CELL IS `B5`'s
     BLIND SPOT RATHER THAN AN OVERFLOW. A `.vz-c-actions` with no `grid-area` is
     auto-placed — it still renders, in a position nobody chose, at **zero page overflow**
     and with every test green. `§E6d` caught exactly that on `.vz-table--seats`, and
     `§E4` constraint 7 is the rule: read the computed grid areas, not the overflow number.
     ⚠ The column is conditional on the reader's `edit` grant, so the row has three areas
     for some people and four for others; naming `actions` on its own line means the card
     is correct either way rather than only for the reader who has the link. */
  .vz-table--languages tr {
    grid-template-areas:
      'name    status'
      'detail  detail'
      'actions actions';
  }

  .vz-table--languages .vz-c-name   { grid-area: name; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  .vz-table--languages .vz-c-status { grid-area: status; justify-self: end; text-align: right; }

  /* ⛔ NOT `display: flex` ON THE CELL — `§E6e`'s finding, pinned by
     `testNoTableCellIsTakenOutOfTheTableBoxModel`: a `<td>` set to flex stops being a
     table-cell. Here the card layout has already made every cell a block, and one link
     needs no row container at all. */
  .vz-table--languages .vz-c-actions { grid-area: actions; width: auto; white-space: normal; }

  .vz-table--languages .vz-c-detail {
    grid-area: detail;
    width: auto;
    white-space: normal;
  }

  /* 🔴 THE CONFIG TABLE IS A ONE-COLUMN CARD, AND THAT IS `§6` ITEM 64 AVOIDED
     RATHER THAN MET. Its second cell holds a VALUE — and one of them,
     `billing.supplier_address`, is a three-line postal address. In a
     `minmax(0, 1fr) auto` card that cell takes the whole width and the first
     column computes to ZERO, with the key spilling across it and page overflow,
     the wrapper and every test reading clean. `§Q5` reproduced exactly that and
     the cause was a sentence in a table cell.

     ▶ So the key sits ABOVE its value, one per line, which is the per-table
     answer that finding names. `242px 0px` is the correct reading for a
     deliberate one-column card — `.vz-table--drift`, this batch's known-good
     one, measures the same. */
  .vz-table--config tr {
    grid-template-columns: minmax(0, 1fr);
    grid-template-areas:
      'name'
      'detail';
  }

  .vz-table--config .vz-c-name { grid-area: name; }

  /* 🐛 🔴 MEASURED, NOT ANTICIPATED — AND IT IS `§6` FINDING 10 IN ITS PUREST FORM.
     The `§E6` sweep read **13 overflows at 320**, cascading from one cell through the
     table, the wrapper, the panel and the DOCUMENT (+7px). The cause was two setting
     keys: `billing.payment_reminder_banner_days` has a **min-content width of 273px
     in a 212px cell**, and `billing.free_occurrence_threshold` 244px. A dotted setting
     key is a machine-written string with no break opportunity in it — a migration
     filename by another name, which is the string finding 10 was written for.

     ▶ PROVED IN BOTH DIRECTIONS IN THE BROWSER BEFORE THIS RULE WAS WRITTEN
     (`§6`, from Q9): 13 overflows → **0** with this rule injected → 13 again with it
     removed. ⛔ AND ONE FEWER COLUMN COULD NOT HAVE HELPED, which is why this batch's
     usual answer is the wrong one here: the card is ALREADY one column, so the token
     is wider than the whole card and no arrangement of cells changes that.

     ⚠ ON THE PAIRING AND NEVER ON `.vz-mono`, which is finding 10 verbatim: that class
     is worn by member codes and phone numbers, which must NOT break. A setting key is
     read off a screen; a member code is dictated down a phone. */
  .vz-table--config .vz-c-name .vz-mono {
    overflow-wrap: anywhere;
  }

  .vz-table--config .vz-c-detail {
    grid-area: detail;
    width: auto;
    /* `pre-line` here too, for the reason the desktop rule below carries — the card must
       not be the width the desktop table is honest at. */
    white-space: pre-line;
  }
}

/* 🐛 🔴 A STORED NEWLINE IS PART OF THE VALUE, AND THE PANEL CLAIMS TO SHOW IT.
   `billing.supplier_address` holds three lines. HTML collapses them, so the text walk
   read *"Covai Tech Park, Site No: 86, 87 Kovai Thiru Nagar, Kalapatty (East)…"* under a
   caption promising the values **as stored** — a small lie, and the kind `§6` finding 12
   catalogues: a sentence on the page that the page itself contradicts.

   ⚠ `pre-line` AND NOT `pre`, WHICH IS THE DISTINCTION `.vz-audit-values` DRAWS FROM THE
   OTHER SIDE. That class is `white-space: pre` because *"a value split across two lines is
   a value somebody copies wrongly"* — right for a JSON blob. This is a postal address: the
   newlines are meaning, and the runs of spacing are not, so `pre-line` keeps the first and
   collapses the second.

   ⛔ AND NOT `nl2br()` IN THE VIEW. `PHASE_Q §Q-B` item 9 settled that staff-authored text
   in this portal is escaped plain text; turning stored newlines into markup would be that
   decision taken again, quietly, in a place nobody would look for it. */
.vz-table--config .vz-c-detail {
  white-space: pre-line;
}


/* ===========================================================================
   The break-glass register (Q8)
   ---------------------------------------------------------------------------
   TWO MODIFIERS, NO NEW CELL CLASS. `.vz-c-when`, `.vz-c-name`, `.vz-c-detail`,
   `.vz-c-status`, `.vz-c-role` and `.vz-c-verb` were all in this file before this
   row and every one was GREPPED before it was typed — `§6`'s standing note that
   this batch has added exactly two cell classes in ten rows, and that an invented
   class leaves the base element drawing itself, so it MEASURES AS 0px AND LOOKS
   CLEAN.

   🔴 ⚠ ONE FEWER COLUMN, BEFORE THE SWEEP RATHER THAN AFTER IT — the sixth time
   in this batch, and the first time it was designed in rather than measured out.
   The register carries SIX facts and renders FOUR cells: the actor folds into the
   start time, because a register is read as "who, and when"; and the elevation
   folds into the state, because `masked` and `elevated` describe the ACCESS
   rather than the session — which is the same structural fold Q7 made when it put
   the audience inside the title cell and kept `Window` as a column because a
   sortable column has to be a column. Neither fold is cosmetic: six columns is
   what `§6` finding 7 has measured going wrong five times running.
   =========================================================================== */

/* The reason is prose somebody wrote in a hurry under pressure, so it wraps
   rather than truncating. ⚠ A `reason` is `VARCHAR(255)` and `0016` calls it a
   ticket ref OR a justification — the second of those is a sentence, and a
   register that elided it would hide the one column the whole table exists to
   make reviewable after the fact. */
.vz-table--breakglass .vz-c-detail {
  white-space: normal;
}

/* 🐛 🔴 THREE OF THIS TABLE'S FOUR CELLS HOLD PROSE, AND TWO OF THEM WEAR CLASSES
   BUILT FOR TOKENS. `.vz-c-when` is `nowrap` product-wide because a DATE split
   across two lines reads badly, and `.vz-c-status` is `nowrap` because it
   normally holds one badge. Here the first also carries the member of staff and
   the second carries a badge plus two explanatory sentences — *"Ran out …, and no
   closure was recorded"* and the elevation — so the global assumption is simply
   false for this register.

   🔴 ⚠ AND "ONE FEWER COLUMN" IS THE WRONG ANSWER HERE, WHICH IS WORTH SAYING
   BECAUSE IT HAS BEEN THE RIGHT ONE FIVE TIMES IN THIS BATCH. Measured at 775 —
   `§6`'s worst band, where the sidebar has appeared and the content column is
   narrowest — the table read **684px inside a 436px wrapper, 248px over**, with
   columns of **155 · 97 · 83 · 348**. The probe was run in the browser before a
   line of this was written, `§6`'s rule from Q9:

     · as shipped ................................. 684  (248 over)
     · access cell allowed to wrap ................ 502  ( 66 over)
     · access AND when allowed to wrap ............ 436  (  0 over)

   ▶ SO THE 248px IS ENTIRELY THE TWO `nowrap` DECLARATIONS, and the four columns
   are the minimum a compliance register can carry: who and when, which operator,
   why, and what access. Folding any of them loses the thing the table exists for
   — Q7's test for whether a merge is structural or cosmetic, answered the other
   way. ⚠ AND THE DOCUMENT READ **-15px** THROUGHOUT, which is finding 7 exactly:
   a `.vz-table-wrap` scrolls with its last column off-screen while the page sweep
   reports clean, so the wrapper has to be swept separately. */
.vz-table--breakglass .vz-c-when,
.vz-table--breakglass .vz-c-status {
  white-space: normal;
}

/* ⚠ THE OPERATOR CODE IS A MACHINE-WRITTEN STRING AND IT PAIRS WITH `.vz-mono`,
   which is finding 10's rule and NOT a rule on `.vz-mono` itself — that class is
   worn by member codes and phone numbers, which must not break. `customer_id` is
   a ten-digit token in a narrow cell; the same shape as the migration filename
   that pushed a whole page 145px sideways in Q3, and as the dotted setting key
   that cascaded 13 overflows to the document in Q10. */
.vz-table--breakglass .vz-c-name .vz-mono {
  overflow-wrap: anywhere;
}

@media (max-width: 64rem) {
  /* 🔴 ⚠ EVERY CELL THE VIEW RENDERS IS NAMED HERE — `§6` findings 8, 9 and 11.
     `.vz-c-actions` carries `grid-area: actions` GLOBALLY, so a modifier whose
     areas did not name a cell it renders would drop that cell into an IMPLICIT
     column: measured in Q4 as four tracks where the template named two, with page
     overflow, the wrapper and every test reading clean. ⚠ This table renders no
     actions cell at all — there is no act to offer — so no `actions` area is
     named, and an area named with no cell would collapse harmlessly but would
     claim this register does something it does not.

     🐛 🔴🔴 A ONE-COLUMN CARD, AND ITEM 64 IS WHY — REPRODUCED HERE, NOT INHERITED.
     The first draft paired the last two cells as `'when access'`, and at 320 the
     LAPSED row measured `gridTemplateColumns: 0px 200px`: its access cell holds a
     badge plus TWO sentences — *"Ran out …, and no closure was recorded"* and the
     elevation — so it took the whole 212px card and starved the first track to
     ZERO. The date then overflowed its own track and **overlapped the badge by
     22px**, with the page sweep at 0, the wrapper at 0 and every test green. The
     other two rows were fine, which is what makes it item 64 rather than a width:
     it depends on the CONTENT of one row.

     ▶ SO EVERY CELL GETS ITS OWN LINE. `§6` finding 8 names this as the per-table
     answer, and `.vz-table--config` took it one row earlier for the same reason.
     ⚠ AND FINDING 9's REFUSAL RULE IS WHAT CAUGHT IT: a `0px` TRACK is not itself
     the defect — a deliberate one-column card reads `212px 0px` and is correct —
     so the check is the CELL's `clientWidth`, and here it was the pair of bounding
     rectangles overlapping that proved it. */
  .vz-table--breakglass tr {
    grid-template-columns: minmax(0, 1fr);
    grid-template-areas:
      'name'
      'reason'
      'when'
      'access';
  }

  .vz-table--breakglass .vz-c-name   { grid-area: name; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--breakglass .vz-c-detail { grid-area: reason; white-space: normal; }
  /* ⚠ BOTH START-ALIGNED NOW THAT EACH HAS ITS OWN LINE. `justify-self: end` on the
     access cell was there to hold the right-hand half of a two-column card; kept,
     it would ragged-right a full-width cell against a left-aligned one above it. */
  .vz-table--breakglass .vz-c-when   { grid-area: when; width: auto; justify-self: start; text-align: left; }
  .vz-table--breakglass .vz-c-status { grid-area: access; width: auto; justify-self: start; text-align: left; }
}

/* ===========================================================================
   The switch (owner, 2026-10-14: "a toggle type instead of a link with on text")
   ---------------------------------------------------------------------------
   🔴 **`aria-pressed` IS THE STYLE HOOK, NOT A SECOND CLASS.** The state a screen reader
   announces and the state an eye sees are the same attribute, so they cannot drift — the
   dashboard's `[draggable="true"]` argument applied to a control rather than an affordance.

   ⚠ **IT IS A `<button>` BECAUSE THIS PAGE SHIPS NO JAVASCRIPT.** A real `<input type=
   checkbox>` would need a Save button or an onchange handler; the button posts the state it
   wants and the track is drawn around it.

   ⚠ AND IT INHERITS NOTHING FROM `.vz-action` ON PURPOSE: a switch is not a button-shaped
   thing with a label, it is a state with a word beside it.
   =========================================================================== */
.vz-switch {
  display: inline-flex;
  align-items: center;
  gap: var(--vz-space-2);
  min-height: var(--vz-tap-min);
  padding: 0;
  border: 0;
  background: none;
  color: var(--vz-ink);
  font: inherit;
  font-size: var(--vz-fs-sm);
  cursor: pointer;
}

/* ⚠ THE TRACK IS `flex: none` SO IT NEVER SHRINKS BESIDE A LONG WORD — *"Waiting for Meta"*
   is three words, and a track that narrowed to fit would read as a different control. */
.vz-switch-track {
  flex: none;
  position: relative;
  display: inline-block;
  width: 2.25rem;
  height: 1.25rem;
  box-sizing: border-box;
  border: 1px solid var(--vz-border-strong);
  border-radius: var(--vz-radius-pill);
  background: var(--vz-surface-2);
  transition: background-color .15s ease, border-color .15s ease;
}

.vz-switch-knob {
  position: absolute;
  top: 1px;
  inset-inline-start: 1px;
  width: 1rem;
  height: 1rem;
  border-radius: 50%;
  background: var(--vz-surface);
  box-shadow: 0 1px 2px rgb(0 0 0 / 25%);
  transition: inset-inline-start .15s ease;
}

/* 🔴 **ON IS THE ARIA STATE.** No `.is-on`, nothing for a view to forget. */
.vz-switch[aria-pressed="true"] .vz-switch-track {
  border-color: var(--vz-primary-strong);
  background: var(--vz-primary-strong);
}

.vz-switch[aria-pressed="true"] .vz-switch-knob {
  inset-inline-start: calc(100% - 1rem - 1px);
}

/* ⚠ **THE WORD STAYS, AND IT IS NOT DECORATION.** A switch with no label is a control whose
   state is carried by colour and position alone — which `.vz-badge--*`'s own comment calls
   "not a distinction for everybody". The owner asked for a toggle INSTEAD OF a link, not
   instead of the word. */
.vz-switch-text { font-weight: var(--vz-fw-medium); }

.vz-switch:hover .vz-switch-track { border-color: var(--vz-primary); }

.vz-switch:focus-visible {
  outline: 2px solid var(--vz-primary);
  outline-offset: 2px;
}
/* ===========================================================================
   The notification manager (owner, 2026-10-14)
   ---------------------------------------------------------------------------
   🔴 **FOUR COLUMNS, AND THE TWO TICK COLUMNS ARE `.vz-c-pick`** — the class this product
   already uses for a column whose whole content is a control, centred by
   `.vz-table .vz-c-pick`. ⚠ The `<th>` carries it too, which is `F38`'s pairing rule and what
   `testTheTickColumnIsCentredWhereItCanWinAndItsHeaderCarriesIt` asserts: a header without the
   class leaves the heading left while the cells below it are centred.

   ⚠ **THE NAME CELL WRAPS AND ITS SUB-META STACKS**, like `.vz-table--templates` two hundred
   lines up. A notification's purpose is a sentence, and `white-space: nowrap` on a sentence is
   what `§Q2` measured as 133px of sideways scroll inside a wrapper reading zero overflow.
   =========================================================================== */
.vz-table--notifications .vz-c-name {
  white-space: normal;
}

.vz-table--notifications .vz-c-name .vz-sub-meta {
  display: block;
}

/* ⚠ AND THE HINT UNDER A TICK STACKS TOO — *"no template, so it cannot be switched on"* is a
   sentence, and beside a one-word badge it would otherwise sit on the same line and double the
   column's width. */
.vz-table--notifications .vz-c-pick .vz-sub-meta {
  display: block;
}

/* 🔴 **A TWO-COLUMN CARD BELOW 64rem, AND THE COLUMN COUNT IS RESTATED ON PURPOSE.**
   `.vz-table tr` is `minmax(0, 1fr) auto`; naming AREAS against that track list without
   restating it leaves the second track in place for auto-placement to find — `§E6d`'s defect,
   measured as four tracks where the template names two. `.vz-table--breakglass-roles` and
   `.vz-table--config` carry the same restatement and the longer version of this warning.

   ⚠ **THE TWO CHANNELS SIT SIDE BY SIDE RATHER THAN STACKING.** They are the row's two
   controls and reading them as a pair is the whole point of the screen; stacking would make a
   card four lines tall and put *"Email: On"* and *"WhatsApp: Off"* two scroll-lengths apart on
   a phone. */
@media (max-width: 64rem) {
  .vz-table--notifications tr {
    grid-template-columns: minmax(0, 1fr) minmax(0, 1fr);
    grid-template-areas:
      'name name'
      'who  who'
      'mail chat';
  }

  .vz-table--notifications .vz-c-name { grid-area: name; }
  .vz-table--notifications .vz-c-kind { grid-area: who; white-space: normal; }

  /* 🐛 ⚠ **A CLASS PER CELL, BECAUSE `:nth-of-type()` CANNOT PICK ONE.** The first draft used
     `.vz-c-pick:nth-of-type(3)`, and `StylesheetStaticTest` refuses it by name: the pseudo-class
     counts `<td>` ELEMENTS among siblings, not cells carrying a class, so it matches nothing
     and the cell falls through to its unscoped `grid-area`. ▶ `.vz-c-money`/`.vz-c-total` are
     the pair already in this vocabulary doing the same thing. */
  .vz-table--notifications .vz-c-mail { grid-area: mail; text-align: start; }
  .vz-table--notifications .vz-c-chat { grid-area: chat; text-align: start; }
}

/* 🔴 A DELIBERATE ONE-COLUMN CARD, AND ITEM 64 IS WHY IT IS SPELLED OUT.
   `.vz-table tr` is `minmax(0, 1fr) auto` below 44rem, so a second-column cell
   holding a LIST takes the whole card and the first computes to `0px` — the role
   name spilling across the verbs with page overflow, the wrapper and every test
   reading clean. The verbs here are up to four words joined by separators, which
   is exactly the content that triggers it. ⚠ AND A `0px` TRACK IS NOT ITSELF THE
   DEFECT: `§6` finding 9 says to refuse on a CELL whose `clientWidth` is 0, never
   on the track, because a one-column card legitimately reads `242px 0px` — which
   is what this table does on purpose, the same as `.vz-table--drift` and
   `.vz-table--config` before it. */
@media (max-width: 64rem) {
  /* ⚠ THE COLUMN COUNT IS RESTATED AND THAT IS NOT REDUNDANT. `.vz-table tr` sets
     `minmax(0, 1fr) auto`; naming one-column AREAS against a two-column TRACK list
     leaves the second track in place for auto-placement to find. `.vz-table--config`
     restates it for the same reason, and its comment above is the longer version. */
  .vz-table--breakglass-roles tr {
    grid-template-columns: minmax(0, 1fr);
    grid-template-areas:
      'role'
      'verbs';
  }

  .vz-table--breakglass-roles .vz-c-role { grid-area: role; font-size: var(--vz-fs-md); color: var(--vz-ink); white-space: normal; }
  .vz-table--breakglass-roles .vz-c-verb { grid-area: verbs; width: auto; white-space: normal; text-align: left; }
}

/* ===========================================================================
   The control plane's billing screen (`PHASE_Q §Q-H` row 7)
   ---------------------------------------------------------------------------
   THREE TABLES, AND EVERY ONE BRINGS ITS OWN MODIFIER. The base card grid's
   own comment is the instruction: *"a NEW table must bring its own modifier
   rather than inherit member column names by accident"* — an unnamed cell is
   auto-placed in an order nobody chose and STILL MEASURES ZERO OVERFLOW,
   because it fits. That is B5's blind spot, and `.vz-table--chits`,
   `--schedule`, `--roster`, `--worklist`, `--drift` and `--config` each exist
   because of it.

   ⚠ AND EVERY `grid-area` BELOW IS SCOPED TO ITS MODIFIER, without exception.
   The unscoped `.vz-c-money { grid-area: value }` further up this file is the
   worked example of why: two money cells on one row both fell through to it,
   landed in the same area and stacked. A cell class on a money screen is the
   one most likely to already mean something somewhere else.
   =========================================================================== */

/* THE OPERATOR ROW'S THREE TEXT CELLS ALL WRAP, AND THEY MUST. Each stacks a
   value over two to four `.vz-sub-meta` lines — a billing period, a negotiated
   rate, a place of supply, a discount window — and the base `.vz-table td` rule
   keeps a cell on one line. Without this the plan cell alone is a full sentence
   pushed off the right edge with page overflow reading zero, which is the exact
   shape the drift table's comment describes. */
.vz-table--billing-operators .vz-c-plan,
.vz-table--billing-operators .vz-c-usage,
.vz-table--billing-operators .vz-c-contact,
.vz-table--invoice-lines .vz-c-desc,
.vz-table--invoice-lines .vz-c-qty {
  white-space: normal;
}

/* ⚠ `anywhere` AND NOT `break-all` — C3b's reason, which `.vz-c-ref` and the
   admin `.vz-c-name` rules above both carry: it breaks a word only when there
   is no other opportunity. An operator's name still wraps between its words,
   and an invoice number (`VTC/2026-27/0001`) or an email address — neither of
   which has a space in it — stops being the thing that widens the table. */
.vz-table--billing-operators .vz-c-name,
.vz-table--billing-operators .vz-c-contact {
  overflow-wrap: anywhere;
}

/* The sub-lines under each cell are their own rows, as everywhere else in the
   product — a plan cell reads as mode / status / period / place, one per line,
   rather than as a run-on sentence. */
.vz-table--billing-operators .vz-c-name .vz-sub-meta,
.vz-table--billing-operators .vz-c-plan .vz-sub-meta,
.vz-table--billing-operators .vz-c-usage .vz-sub-meta,
.vz-table--billing-operators .vz-c-contact .vz-sub-meta,
.vz-table--invoice-lines .vz-c-desc .vz-sub-meta,
.vz-table--invoice-lines .vz-c-qty .vz-sub-meta {
  display: block;
}

/* MONEY RIGHT-ALIGNS AND THE COLUMN DOES NOT SHRINK. Two totals in a column
   are only comparable at a glance when their last digits line up, which is the
   whole reason a money column is right-aligned anywhere in this product.
   ⚠ AND `white-space: nowrap` IS NOT DECORATION HERE — C3c shipped a defect in
   exactly this shape: a minus sign wrapped away from its amount, so a negative
   till read POSITIVE on the next line. No figure on this screen is negative
   today; the rule is here so none can be broken tomorrow. */
.vz-table--invoice-lines .vz-c-amount {
  text-align: right;
  white-space: nowrap;
}

/* ⏹ ── `.vz-table--billing-invoices` IS GONE, AND SO ARE ITS NINE RULES (F31) ──
   The overview's five-row invoice glance went behind the register's button, so the
   modifier has no view rendering it. ⛔ `StylesheetStaticTest::
   testNoTableModifierIsDefinedWithoutAViewThatRendersIt` is the guard that says so,
   and it is the reason this was deleted rather than left "in case": a modifier
   painting nothing is indistinguishable from one whose view forgot to name it.
   ⚠ **NOTHING WAS LOST.** `.vz-table--invoice-register` below carries the same four
   cells with widths this one never had, and the two tables were the same table twice
   from the day the register was built. */

/* ── THE INVOICE REGISTER (owner's item 7) ────────────────────────────────
   `/admin/billing/invoices`. It shares the overview table's cell treatments below
   and carries its own widths, because it is the table F12 warned about before it
   was written rather than after.

   ⛔ `table-layout: fixed` FROM THE FIRST COMMIT, WHICH IS F12's LESSON APPLIED
   RATHER THAN QUOTED. On `/admin/jobs` a 503-character cell under `table-layout:
   auto` starved its neighbour to 71px, and the `overflow-wrap: anywhere` below
   then turned the squeeze into mid-word shattering rather than a visible overflow
   — `provisioning_worker` given 41px of the 120px it needs. ▶ This table carries
   the same two ingredients: an invoice number with no spaces in it
   (`VTC/2026-27/0001`) beside operator names of very different lengths. Fixed
   layout takes the decision away from the content, so no row can starve another.

   ⚠ THE WIDTHS SIT ON `thead th` AND NOT ON THE `.vz-c-*` CELLS, AND THAT IS NOT
   A STYLE CHOICE. Under `table-layout: fixed` the browser reads its columns from
   the FIRST ROW ONLY and ignores every later one. F12's fix was first written
   against the `<td>` classes and the tables came back 305/305/305 — fixed layout
   applied, the widths did not. ▶ Every header cell here is built by the view's
   `$sortHeader()` helper and carries no class of its own, so POSITION is what
   there is to aim at.

   ⚠ THE FOURTH COLUMN IS LEFT TO TAKE WHAT REMAINS. Money is the one column
   whose content has a known short maximum, and pinning all four would make the
   table's total width a number this file has to keep correct.

   ⛔ AND IT IS SCOPED TO THE WIDE BAND. Below 64rem every row becomes a card
   (`tr { display: grid }`), where `table-layout` means nothing — a rule applied
   at every width would be inert at the narrow ones and would invite somebody to
   "simplify" it later. */
@media (min-width: 64.0625rem) {
  .vz-table--invoice-register { table-layout: fixed; }

  .vz-table--invoice-register thead th:first-child  { width: 15rem; }
  .vz-table--invoice-register thead th:nth-child(2) { width: 18rem; }
  .vz-table--invoice-register thead th:nth-child(3) { width: 11rem; }
}

/* The register's cells, on the same reasoning as the overview table's above:
   `anywhere` rather than `break-all` so an operator's name still wraps between
   its words and only the invoice number — which has no space in it — is broken
   when there is no other opportunity. */
.vz-table--invoice-register .vz-c-invoice-no,
.vz-table--invoice-register .vz-c-org {
  overflow-wrap: anywhere;
}

.vz-table--invoice-register .vz-c-invoice-no .vz-sub-meta,
.vz-table--invoice-register .vz-c-org .vz-sub-meta,
.vz-table--invoice-register .vz-c-status .vz-sub-meta,
.vz-table--invoice-register .vz-c-total .vz-sub-meta {
  display: block;
}

/* MONEY RIGHT-ALIGNS AND DOES NOT SHRINK — the overview table's rule and its
   reason: two totals are comparable at a glance only when their last digits line
   up, and C3c shipped a minus sign that wrapped away from its amount, so a
   negative till read POSITIVE on the next line. */
.vz-table--invoice-register .vz-c-total {
  text-align: right;
  white-space: nowrap;
}

/* ⚠ THE TAX SUB-LINE IS NOT BOLD AND THE TOTAL IS. They sit in one cell, and
   without this the smaller figure reads with the same weight as the one being
   compared down the column. */
.vz-table--invoice-register .vz-c-total .vz-sub-meta {
  font-weight: 400;
  color: var(--vz-subtle);
}

.vz-table--invoice-lines .vz-c-qty {
  text-align: right;
}

/* THE ONE WRITE ON THE SCREEN, INSIDE A TABLE CELL. It is a form rather than a
   field because it posts on its own, and it wraps because a 255-character
   address box and a button do not fit side by side in a table column at any
   width this product supports. */
.vz-inline-form {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2);
  align-items: center;
  margin-top: var(--vz-space-2);
}

/* 🐛🔴🔴 `.vz-control`, AND THE FIRST DRAFT SAID `.vz-input` — WHICH IS A CLASS
   WITH NO BASE RULE IN THIS PRODUCT, AND WRITING THIS RULE IS WHAT HID THAT.

   `OPERATING_MODE §7` item 51: *"`.vz-input` is a class with no CSS rule anywhere
   in the product… the real component is `.vz-control`, which carries
   `min-height: var(--vz-tap-min)`, the border, the radius and the focus ring — so
   anything classed `.vz-input` renders as a bare browser control at whatever
   height the user agent picks."* Measured on the dashboard row: **21px against
   `.vz-control`'s 44**, under item 19's tap-target floor.

   🔴🔴 AND THE TRAP IS SHARPER THAN REUSING THE WRONG NAME. `StylesheetStaticTest`
   reports a class *"rendered and painted by nothing"* — so `.vz-input` was
   catchable **only while it had no rule at all**. ▶ Adding `.vz-inline-form
   .vz-input { … }` gave it its first rule in the product's history and **silenced
   the guard**, while leaving the control still missing its height, border and
   focus ring. **A partial rule on an unstyled class is worse than none: it buys
   the appearance of styling and the silence of the test.**

   ⚠ THE FIX IS THE COMPONENT, NOT A BETTER RULE — `StylesheetStaticTest`'s own
   message says so: *"either add the rule or use the component that already
   exists."* Both hand-written inputs on this screen are `.vz-control` now, and
   this rule only makes one lay out inside a flex row. */

/* 🐛🔴 `min-width: 0`, AND THE FIRST DRAFT SAID `12rem` — WHICH WAS BOTH THE
   OVERFLOW ABOVE AND A `StylesheetStaticTest` FAILURE.

   `testEveryGrowableFlexItemEitherCanShrinkOrSaysWhyNot` is `§E25` finding 10's
   guard: a flex item that GROWS must be able to SHRINK below its own content, or
   it pushes its container sideways in the viewport band nobody measures. ▶ It
   fired on this rule, and it was right — the 12rem floor is exactly what took
   the operators table 587px inside a 461px wrap at 785px.

   ⚠ AND THE FLOOR IS NOT NEEDED NOW THAT THE TABLE CARDS AT 64rem. Below that
   the contact cell is a full-width card row, and above it the viewport is 1024px
   or wider with four columns — so the input has room in both, and
   `flex-basis: 12rem` still asks for a sensible size without demanding one.
   ⚠ THE `<input>`'s OWN INTRINSIC WIDTH still floors the TABLE COLUMN above
   64rem — `min-width: 0` governs flex shrinking, not min-content contribution —
   which is why the breakpoint is the fix and this is only the half of it that
   stops a growable item refusing to shrink. */
.vz-inline-form .vz-control {
  flex: 1 1 12rem;
  min-width: 0;
}

/* 🐛🔴 THE OPERATOR TABLE BECOMES CARDS AT 64rem, NOT 44rem, AND THAT IS A
   MEASURED FIX RATHER THAN A PREFERENCE.

   Measured through the `§E6` harness: at a 785px viewport the table was **587px
   inside a 461px wrap** — it scrolled sideways with the contact form's button
   off-screen — while `documentElement.scrollWidth - clientWidth` read **0 at
   every width from 320 to 1440**, because `.vz-table-wrap` was doing exactly its
   job. That is row 14's finding, and it is only visible by comparing
   `table.scrollWidth` against `wrap.clientWidth`.

   ⚠ THE CAUSE WAS THIS FILE'S OWN `min-width: 12rem` ON THE CONTACT INPUT. An
   `<input>` contributes an intrinsic min-content width to the table algorithm
   whatever `min-width: 0` says about flex shrinking, so the column had a 232px
   floor out of a 461px table. **A cell holding a form has a floor the content
   cannot argue with**, which is `.vz-c-actions`' lesson from `§8a` row 7i in a
   different costume.

   ▶ SO THE BAND WHERE FOUR CONTENT COLUMNS PLUS A FORM DO NOT FIT — roughly 704
   to 1000px — GETS THE CARD LAYOUT THAT WAS ALREADY BUILT FOR IT. Shrinking the
   floor was tried first and measured: it took the table to 564px in 561px at
   900, still scrolling, because the input's intrinsic width is the floor. ⛔ And
   dropping a column was the other option `.vz-table--operators` took for its
   connection column; here every column is the answer to a different question the
   screen exists to answer. */
@media (max-width: 64rem) {
  /* ⚠ THE DISPLAY SWITCH IS RESTATED, because the base one lives in the 44rem
     block and this table needs it 20rem earlier. Only this modifier is named, so
     no other table's breakpoint moves. */
  .vz-table--billing-operators,
  .vz-table--billing-operators tbody,
  .vz-table--billing-operators tr,
  .vz-table--billing-operators td {
    display: block;
    width: auto;
  }

  /* The header row still exists for assistive technology; it simply has nothing
     to label once the grid is gone. */
  .vz-table--billing-operators thead {
    position: absolute;
    width: 1px;
    height: 1px;
    overflow: hidden;
    clip: rect(0 0 0 0);
    clip-path: inset(50%);
  }

  /* 🐛🔴 A DELIBERATE ONE-COLUMN CARD, AND IT IS THE DRIFT CARD'S DEFECT CAUGHT
     ON THIS ROW RATHER THAN INHERITED FROM IT.

     The first version of this block was `minmax(0, 1fr) auto` with the plan cell
     beside the name. Measured at 375: `grid-template-columns` computed to
     **`0px 255px`** — the operator's name had `clientWidth: 0` and a 620px-tall
     box, its text spilling across the card — while page overflow read **0** and
     every cell still resolved to a correctly *named* area. ▶ Row 14's check
     ("every `td` resolves to a named area") and `§8a` row 7i's sharper one ("read
     the NAMES") both PASSED; only printing the computed width per cell found it.

     ⚠ THE CAUSE IS THE PLAN CELL'S CONTENT, and it is not going to get shorter:
     it stacks the mode, the status, the billing period, a negotiated rate, an
     override reason, the place of supply and a discount window — including
     sentences like *"No place of supply on record — this operator is skipped by
     the invoicing run"*. An `auto` track sized by that takes the whole card.
     `.vz-table--drift` and `.vz-table--config` reached one column the same way.

     ⚠ AND THE COLUMN COUNT IS RESTATED RATHER THAN LEFT TO INHERIT — the base
     `.vz-table tr` rule sets `minmax(0, 1fr) auto`, and naming one-column AREAS
     against a two-column TRACK list leaves the second track in place for
     auto-placement to find. `.vz-table--breakglass-roles` carries the longer
     version of this warning. */
  .vz-table--billing-operators tr {
    display: grid;
    grid-template-columns: minmax(0, 1fr);
    grid-template-areas:
      'name'
      'plan'
      'usage'
      'contact';
    gap: var(--vz-space-1) var(--vz-space-3);
    margin-bottom: var(--vz-space-3);
    padding: var(--vz-space-4);
    border: 1px solid var(--vz-border);
    border-radius: var(--vz-radius-lg);
    background: var(--vz-surface);
  }

  .vz-table--billing-operators tbody tr:hover td { background: none; }

  .vz-table--billing-operators td {
    padding: 0;
    border-bottom: 0;
  }

  .vz-table--billing-operators .vz-c-name    { grid-area: name; font-size: var(--vz-fs-md); }
  /* ⚠ `justify-self: start` AND `text-align: left`, EXPLICITLY. The desktop rule
     for a status-ish column ragged-rights it; on a full-width card that would set
     one block of text against the left-aligned ones above and below it, which is
     the note `.vz-table--breakglass` carries for its own `when` cell. */
  .vz-table--billing-operators .vz-c-plan    { grid-area: plan; justify-self: start; text-align: left; }
  .vz-table--billing-operators .vz-c-usage   { grid-area: usage; }
  .vz-table--billing-operators .vz-c-contact { grid-area: contact; }

  /* ⚠ THE WRAP'S OWN CHROME GOES WITH THE TABLE, and it needs its own class
     because the base reset is keyed to the 44rem block. Between 44 and 64rem the
     wrap would otherwise draw a bordered, scrollable box around a stack of cards
     that are already bordered — two frames around one list. ⛔ `:has()` would
     express it without the extra class; this file already argues against relying
     on it where the fallback is wrong rather than merely plainer, and a
     double-bordered list is exactly that. */
  .vz-table-wrap--cards-early {
    border: 0;
    background: none;
    overflow-x: visible;
  }

  /* 🔴 THE OPERATOR'S OWN SUBSCRIPTION INVOICES (`§Q-R4`, `§7` item 77) — THE SECOND
     TABLE TO NEED THE EARLY BREAKPOINT, AND THE SWEEP IS WHAT PUT IT HERE.

     Measured at the 44rem default: **734px inside a 685px wrap at 1024, 561 at 900
     and 677 at 768** — it scrolled sideways across the whole 768→1150 band, at
     **0px page overflow** the entire time, because `.vz-table-wrap` was doing its
     job. ⚠ THAT BAND IS WHERE THE 240px NAVIGATION IS ALREADY DRAWN (it arrives at
     48rem) AND THE TABLE HAS NOT YET BECOME A CARD — the squeeze is structural, not
     this table's fault, which is why `.vz-table--billing-operators` hit it first.

     ⚠ AND THE FLOOR WAS FIVE `nowrap` COLUMNS, THE WIDEST BEING THE PERIOD at 196px:
     "01 Aug 2026 to 31 Aug 2026" is 26 characters `.vz-c-when` will not break.
     🔴 **BOTH FIXES ARE HERE AND NEITHER WAS SUFFICIENT ALONE, WHICH THE SWEEP HAD TO
     SHOW TWICE.** Letting the period wrap saves ~84px — and the 900px wrap is 561, so
     the table would still have scrolled there. Moving the breakpoint to 64rem fixed
     768→1024 — and left a residual band at **1028–1068**, eleven widths just above it
     where the column is wide enough for the nav but not yet for five rigid columns.
     ▶ The breakpoint is the structural fix; the wrap (with its `min-width` floor, in
     the base block above) is what closes the seam it leaves behind. */
  .vz-table--subscription,
  .vz-table--subscription tbody,
  .vz-table--subscription tr,
  .vz-table--subscription td {
    display: block;
    width: auto;
  }

  .vz-table--subscription thead {
    position: absolute;
    width: 1px;
    height: 1px;
    overflow: hidden;
    clip: rect(0 0 0 0);
    clip-path: inset(50%);
  }

  /* ⚠ THE COLUMN COUNT IS RESTATED RATHER THAN LEFT TO INHERIT — the base
     `.vz-table tr` rule sets `minmax(0, 1fr) auto`, and naming two-column AREAS
     against that track list is fine, but stating it here is what stops a later
     change to the base from silently re-placing these cells.

     ⚠ THE SERIAL LEADS AND THE STATUS SITS OPPOSITE IT: "VTC/2026-27/0007 — Paid"
     is the line somebody scans for. The period takes its own row because it is long
     and has nothing to share one with, and the two money cells share the last row so
     the comparison the card exists for is one glance. */
  .vz-table--subscription tr {
    display: grid;
    grid-template-columns: minmax(0, 1fr) auto;
    grid-template-areas:
      'code   status'
      'period period'
      'total  paid'
      'doc    doc';
    gap: var(--vz-space-1) var(--vz-space-3);
    margin-bottom: var(--vz-space-3);
    padding: var(--vz-space-4);
    border: 1px solid var(--vz-border);
    border-radius: var(--vz-radius-lg);
    background: var(--vz-surface);
  }

  /* 🔴 THE FOURTH ROW IS A DECISION, AND WITHOUT IT THE BROWSER MADE ONE (`§7` item 106).
     Measured at 375px before this rule existed: `.vz-c-document` reported
     `grid-area: auto` and was AUTO-PLACED into an implicit fourth row — it looked
     right, and it looked right by accident. The named areas above are what stops a
     later change to the track list re-placing it somewhere else silently.

     ⚠ IT SPANS BOTH COLUMNS AND SITS `start`. The three rows above pair a label with
     its opposite number — serial against status, total against received — and a
     download is not the counterpart of anything; giving it half the width would
     invite a reader to look for what belongs beside it.

     ⚠ AND AN EMPTY CELL TAKES NO ROW AT ALL. A reader without `subscription.export`
     (`0097`) gets no button and no refusal, and `:empty` is why their cards do not
     each carry a blank fourth row. The view emits nothing rather than a `&nbsp;`
     precisely so this can match. */
  .vz-table--subscription .vz-c-document { grid-area: doc; justify-self: start; }
  .vz-table--subscription .vz-c-document:empty { display: none; }

  .vz-table--subscription tbody tr:hover td { background: none; }

  .vz-table--subscription td {
    padding: 0;
    border-bottom: 0;
  }

  .vz-table--subscription .vz-c-code   { grid-area: code; font-size: var(--vz-fs-md); color: var(--vz-ink); }
  /* 🐛 F154 (found driving pricing part 3, 2026-11-15): AT 375px THE SERIAL RAN UNDER THE STATUS CHIP.
     `.vz-c-code` is `nowrap` for the table, so in the card's shrinking first column a 16-character mono
     serial beside the widest chip ("Nothing to pay", 125px) overflowed by 26px and "VTC/2026-27/0003"
     read as "VTC/2026-27/000". ▶ On the card it may wrap — and only after a slash, where the view puts
     a `<wbr>` — so a serial breaks into its series and its number, never through a number. */
  .vz-table--subscription .vz-c-code   { white-space: normal; }
  .vz-table--subscription .vz-c-status { grid-area: status; justify-self: end; }
  /* ⚠ `white-space: normal` ON THE CARD. `.vz-c-when` is `nowrap` for the table, which
     is right there and wrong on a 320px card where 26 unbreakable characters are most
     of the width. The dates still read as one phrase because the row is their own. */
  .vz-table--subscription .vz-c-when   { grid-area: period; white-space: normal; color: var(--vz-subtle); }
  .vz-table--subscription .vz-c-money  { grid-area: total; color: var(--vz-ink); }
  .vz-table--subscription .vz-c-paid   { grid-area: paid; justify-self: end; }
}

@media (max-width: 64rem) {
  /* ⏹ THE OVERVIEW'S INVOICE CARD WENT WITH ITS TABLE (F31) — five rules, the
     last of the `.vz-table--billing-invoices` modifier. See the note beside the
     cell treatments above for why a dead modifier is deleted rather than kept. */

  /* THE REGISTER'S CARD LEADS WITH THE NUMBER AND THE MONEY, because those are
     the pair somebody scanning a list of documents is matching against.
     ⏸ **THIS COMMENT USED TO END *"the same order the overview's card uses, so the
     two screens do not disagree about what an invoice looks like"*** — true, and
     F31 removed the screen it was agreeing with. ▶ The order is unchanged; what
     went is the second card it was keeping faith with.

     ⚠ THE STATE TAKES A ROW OF ITS OWN RATHER THAN RIDING BESIDE THE TOTAL.
     It carries a badge AND, on an unpaid invoice, a due date; squeezed into a
     half-width area beside the money it would wrap to two lines on a phone and
     push the operator's name down with it. */
  .vz-table--invoice-register tr {
    grid-template-areas:
      'no     total'
      'org    org'
      'status status';
  }

  .vz-table--invoice-register .vz-c-invoice-no { grid-area: no; font-size: var(--vz-fs-md); }
  .vz-table--invoice-register .vz-c-total      { grid-area: total; justify-self: end; }
  .vz-table--invoice-register .vz-c-org        { grid-area: org; }
  .vz-table--invoice-register .vz-c-status     { grid-area: status; }
  /* THE LINE CARD PUTS THE DESCRIPTION ON ITS OWN ROW. An invoice line's text
     is a whole sentence — *"Vazhai Chit subscription, 23 Aug to 31 Aug — 5
     member-occurrence(s) recorded, none billable"* — and it is the longest
     string this screen renders. Sharing a track with the amount is precisely
     what starves the first column to zero. */
  .vz-table--invoice-lines tr {
    grid-template-areas:
      'desc   desc'
      'qty    amount';
  }

  .vz-table--invoice-lines .vz-c-desc   { grid-area: desc; color: var(--vz-ink); }
  .vz-table--invoice-lines .vz-c-qty    { grid-area: qty; text-align: left; }
  .vz-table--invoice-lines .vz-c-amount { grid-area: amount; justify-self: end; }

  /* ⏹ `.vz-table--billing-tiers` went in pricing part 7 with the occurrence price list it laid out. */
}

/* ===========================================================================
   The operator notification centre (§Q-R12, OPERATING_MODE §7 item 81)
   ---------------------------------------------------------------------------
   Five classes and one modifier, and every one of them was GREPPED across
   `app/Views` and this file before it was typed — `§8a` row 7i's finding is
   that reusing a name another module already styles wins on source order and
   MEASURES AS 0px while the card is wrong, and inventing one leaves the base
   element drawing itself, which also measures clean.

   ⚠ NOT A `.vz-table`. A broadcast is a title and a paragraph of prose, not a
   row of fields — `§7` item 63's eleven registers are the argument for not
   making it a twelfth, and there is no column here anybody would sort on.
   =========================================================================== */

/* ⚠ A GRID RATHER THAN MARGINS ON THE CARDS, so the gap between two notices is stated once
   and a single notice carries no trailing space.

   🐛 THIS SENTENCE WAS FALSE IN BOTH HALVES FROM THE DAY IT WAS WRITTEN UNTIL `§7` ITEM 116
   (2026-09-18). The card inherited `margin-bottom: 16px` from the form notice it shares a
   class name with, so the measured gap was **28px** (this 12px plus that 16) and a single
   card carried **16px** of trailing space. ▶ The fix is on the card below, which now states
   every box property it cares about. **The claim is true now**; it is left here rather than
   softened because a comment arguing for its own absence is what made the defect findable. */
.vz-notices {
  display: grid;
  gap: var(--vz-space-3);
}

/* 🐛🔴 CONTAINER-SCOPED, AND THE BARE SELECTOR WAS A PRODUCT-WIDE DEFECT (`§7` item 110).
   This card was written as `.vz-notice` — a name the FORM notice already owns, some 7,400
   lines above under its own `--- Notices ---` heading, with `--error` beside it and `--warn`
   further down. Both were `(0,1,0)`, this one is later, so it won `padding`, `border` and
   `background` on every notice in the product and **every warning and every validation
   summary rendered as a plain white panel**: 44 view files and 51 occurrences, measured on
   closing, plus the neutral explainer on `chits/wizard.php` and `members/nominees.php`.
   ▶ The modifiers were correct all along; only their specificity was wrong, so the fix is the
   container this component already sits inside and never a rename.

   ⚠ THE COMMENT ABOVE THIS BLOCK SAYS EVERY CLASS WAS GREPPED BEFORE IT WAS TYPED, AND THAT
   IS STILL TRUE OF THE FIVE IT NAMES — `.vz-notice` itself is the one that was not, because
   it reads as the component's own base name rather than as a borrowed one. That is `§8a` row
   7i's finding arriving through the one class the rule was written about.
   ⛔ `StylesheetStaticTest::testNoClassIsSilentlyRedefinedByASecondTopLevelRule` is what
   notices the next one; nothing in this repository did before item 110.

   🐛🔴 AND THE COLLISION LEAKED THE OTHER WAY TOO — `§7` ITEM 116, FIXED HERE (2026-09-18).
   Item 110 scoped what the CARD was wrongly winning. It never measured what the card was
   wrongly INHERITING: the base rule at `module.css` 880 still matches this `<article>`, and
   three of its declarations were not re-stated here, so they still applied.
     · `margin-bottom: var(--vz-space-4)` — 16px, on a component whose container is a GRID
       specifically so that it would not need one. MEASURED: the gap between two cards was
       **28px** (12px grid gap PLUS the 16px margin) and a single card carried **16px** of
       trailing space inside its own grid. ▶ The comment four lines above this block —
       *"a grid rather than margins on the cards, so the gap is stated once and a single
       notice carries no trailing space"* — **was false in both halves, and had been since
       the component shipped.** A comment arguing for its own absence is the sharpest kind of
       tell, and nothing measured it for two rows.
     · `font-size: var(--vz-fs-sm)` — 13px, so the announcement BODY read a size nothing
       chose. `.vz-notice-body` states no size and the panel prose around it is body copy;
       `§Q-R12`'s *"an operator READING a notice"* argument is about the font FAMILY.
     · `color: var(--vz-muted)` — ⛔ DELIBERATELY NOT FIXED. `-title`, `-body` set
       `--vz-ink`, `-when` sets `--vz-muted`, and `-marks` holds only spans that set their
       own, so **no text in this card inherits it.** A leak with no reader is not a defect,
       and saying so here stops the next reader re-finding it.
   ⚠ SO THIS BLOCK NOW STATES EVERY BOX PROPERTY IT CARES ABOUT and inherits none of another
   component's. ⛔ NOT a scope on line 880: that rule is the base for 44 view files, and
   narrowing it to fix a two-card screen is item 110's cost/benefit inverted.
   ⚠ `-marks` AND `-when` WERE CHECKED AFTER, as the item asked: `-when` is `--vz-fs-xs`
   explicitly and `-marks` holds only badges and `-when`, all with their own sizes — so the
   font change reaches the BODY and nothing else. */
.vz-notices .vz-notice {
  margin: 0;
  padding: var(--vz-space-3);
  border: 1px solid var(--vz-border);
  border-radius: var(--vz-radius-md);
  background: var(--vz-surface);
  font-size: var(--vz-fs-base);
}

/* ⚠ QUIETER, NOT SMALLER. A notice under *Earlier* is still something the reader may have
   come here specifically to re-read — `§7` item 81's answer is that "the notices are still
   reachable" — so it keeps its full wording and loses only its emphasis.

   🐛 SCOPED TOO, AND NOT FOR TIDINESS — THIS IS THE NEIGHBOUR ITEM 110's FIX WOULD HAVE
   BROKEN. Bare, it is `(0,1,0)` and the rule above became `(0,2,0)`, so scoping only the base
   would have handed `background: var(--vz-surface)` to *Earlier* as well and quietly deleted
   the one thing that distinguishes the two panels. Caught before it shipped by asking what
   the raised specificity now outranks, and held afterwards by
   `testThePastNoticeStillOutranksTheCardItVaries`. */
.vz-notices .vz-notice--past { background: var(--vz-surface-2); }

/* The badges and the date, on one line that wraps as a whole rather than mid-badge. */
.vz-notice-marks {
  display: flex;
  flex-wrap: wrap;
  align-items: center;
  gap: var(--vz-space-2);
  margin: 0 0 var(--vz-space-2);
}

.vz-notice-when {
  color: var(--vz-muted);
  font-size: var(--vz-fs-xs);
}

/* ⚠ `<h3>` BECAUSE THE PANEL TITLE ABOVE IT IS AN `<h2>`, and this is the one screen in the
   module that stacks a heading inside a panel — a jump to `<h4>` or back to `<h2>` is the
   outline defect `§D11a` found in a `<caption>`.

   🐛 SCOPED, AND IT IS `.vz-notice`'s DEFECT ONE COMPONENT ALONG (`§7` item 110). The form
   notice has a title rule of its own — `font-weight: var(--vz-fw-bold)` and no `font-size`,
   so it takes the notice's `--vz-fs-sm` — and this one was overriding it on all 13 views
   that render a `<p class="vz-notice-title">`: a `--vz-fs-md`, medium-weight, `--vz-ink`
   line where a bold body-sized one was designed. ⚠ **Item 110 named `.vz-notice` alone**;
   this was found by the scan the item asked for, which is the argument for writing the guard
   before the fix rather than after it. */
.vz-notices .vz-notice-title {
  margin: 0 0 var(--vz-space-2);
  font-size: var(--vz-fs-md);
  font-weight: var(--vz-fw-medium);
  line-height: var(--vz-lh-tight);
  color: var(--vz-ink);
}

/* 🔴 THE AUTHOR'S PARAGRAPHS SURVIVE, AND THIS IS DELIBERATELY NOT `.vz-wording`.
   That class does the same job in `/admin` and is MONOSPACE, because it exists for a
   reviewer checking a template body — the same distinction `.vz-audit-values` draws from the
   other side, where `white-space: pre` is right for a JSON blob and wrong for prose. This is
   an operator READING a notice, so it is the body font.

   ⚠ `pre-wrap` AND NOT `pre-line`: `PHASE_Q §Q-B` item 9 settled that this text is escaped
   plain text with no Markdown renderer, so the author's own line breaks and indentation are
   the only structure the body has. ⛔ AND NOT `nl2br()` IN THE VIEW — that would be the same
   decision taken again, quietly, somewhere nobody would look for it.

   ⚠ `overflow-wrap: anywhere` IS NOT BELT AND BRACES: an incident notice carries a status-page
   URL, which is exactly the shape that pushed a whole page 145px sideways in Q3 with no table
   involved. `anywhere` rather than `break-all` — it only breaks a word when there is no other
   opportunity, so ordinary prose still wraps at its spaces. */
.vz-notice-body {
  margin: 0;
  color: var(--vz-ink);
  line-height: var(--vz-lh-base);
  white-space: pre-wrap;
  overflow-wrap: anywhere;
}

.vz-notice-meta {
  margin: var(--vz-space-2) 0 0;
  color: var(--vz-muted);
  font-size: var(--vz-fs-sm);
}

/* 🔴 THE SENTENCE THAT EXPLAINS WHAT A CLICK LEAVES BEHIND — or why there is no click.
   It is `--vz-ink` rather than `--vz-muted`: the two acts on this screen look identical from
   the outside and only one of them is evidence `0039` keeps, so the difference is the
   message rather than a footnote to it. */
.vz-notice-why {
  margin: var(--vz-space-3) 0 0;
  color: var(--vz-ink);
  font-size: var(--vz-fs-sm);
}

/* ⚠ A `<form>`, WHICH IS WHY THIS EXISTS AT ALL — a form is `display: block` and would
   otherwise put the button on a line of its own with a browser's default margin. */
.vz-notice-acts {
  display: flex;
  flex-wrap: wrap;
  gap: var(--vz-space-2);
  margin: var(--vz-space-3) 0 0;
}

/* 🔴 THE PINNED BANNER (§Q-R12). It is a `.vz-callout` in every visual respect — tone, glyph,
   layout — and this class exists for TWO reasons that are not decoration:

   1. ⚠ IT SEPARATES CONSECUTIVE BANNERS. Two critical broadcasts can be live at once and
      `.vz-callout`'s own margin is a bottom one, so a second would sit flush against the
      first with nothing saying they are two notices rather than one long one.
   2. 🐛 IT GIVES THE BANNER A NAME A TEST CAN SELECT ON. Three assertions in
      `AnnouncementCentreScreenWireTest` measured the wrong thing without it: the BELL prints
      the same title on every screen, so "the banner is pinned" and "the bell is working" are
      indistinguishable by text alone. `§8a` row 7i's rule from the test side — a probe that
      selects by something shared cannot attribute what it finds.

   ⚠ NO TONE OF ITS OWN. `severity` decides that and the three `.vz-callout--*` variants
   already carry it; a colour here would be a second opinion about the same column. */
.vz-notice-banner + .vz-notice-banner { margin-top: var(--vz-space-3); }

/* 🔴 THE READ-ONLY BANNER (2026-10-28) — `TenantReadOnly`. The same two reasons as the notice
   banner above, applied to the banner that sits first: a support-seat or notice banner can
   follow it, and with only a bottom margin between them two notices read as one. And a name a
   test can select on, so "the tenant is read-only" is asserted by THIS element and not by a
   sentence the refusal page also prints. ⚠ No tone of its own — `.vz-callout--warn` has it. */
.vz-read-only-banner + .vz-callout { margin-top: var(--vz-space-3); }

/* 🔴🐛 AN EMPTY HINT MUST NOT TAKE A LINE BOX (owner, 2026-09-29).

   The chit value's hint IS the amount in words, and it is empty until there is an amount —
   so without this an untouched form carries a blank line under the box.

   ⛔ THE FIRST ATTEMPT PUT THE WORDS IN A <p> BESIDE THE FIELD AND SPANNED THE GRID. That
   was wrong at every width and only LOOKED right at the one it was driven at: a sibling of
   the field is its own GRID ITEM, so on the three-column contract step it landed in the next
   free cell — the far left of the following row, under *Group name*, which is where the
   owner found it. The words now live in the field partial's own hint slot, which is inside
   the field's cell by construction and cannot drift to another column. */
.vz-field-hint:empty { display: none; }

/* A money box's amount in words — its own line under the hint, painted like one.
   ⚠ EMPTY UNTIL THERE IS AN AMOUNT, and an empty <p> still takes a line box, which
      would open a gap under every money field nobody has typed in yet. */
.vz-field-words { display: none; }
.vz-field-words:not(:empty) { display: block; }

/* 🔴 PRICING PART 1 — A ONE-COLUMN CARD FOR THE RATE-CARD AND PLAN-HISTORY TABLES (2026-11-14).
   Driven at 375: on the bare `.vz-table` grid (`minmax(0, 1fr) auto`, member column names) the
   labels "BAND" and "CHIT VALUE" printed on top of each other and a date wrapped one word per line.
   The CSS above says a new table brings its own modifier rather than inherit member areas, so
   these four tables stack every cell full width, in source order, each under its own label. */
@media (max-width: 64rem) {
  .vz-table--pricing tr {
    grid-template-columns: minmax(0, 1fr);
    grid-template-areas: none;
  }

  .vz-table--pricing td {
    grid-area: auto;
    justify-self: stretch;
    text-align: start;
    white-space: normal;
  }
}

/* ═══ THIS MONTH SO FAR, IN DETAIL — `/app/subscription` (owner, 2026-11-15) ════════════════════
   The invoice's Annexure A for a month that has not ended. On a desktop it is an ordinary table; on a
   phone a GROUP is a card of labelled lines and the fee, adjustment, GST and total rows
   (`.vz-row-line`) are one line each with the figure on the right.
   ⚠ FOUND IN THE LOCAL DRIVE: as a scrolling `--doc` table the Amount column and the total sat
   off-screen at 375, and at 860 too before the PDF's nine columns came down to six.
   ⚠ `.vz-c-draws` is this table's own: two count columns share a row, and a card needs a grid area
   for each. Its right alignment is the base rule's for every other figure column. */
.vz-table--month-detail .vz-c-draws { text-align: right; }

@media (max-width: 64rem) {
  .vz-table--month-detail tr {
    grid-template-columns: minmax(0, 1fr);
    grid-template-areas:
      'name'
      'value'
      'slots'
      'draws'
      'rate'
      'amount';
  }

  .vz-table--month-detail tr.vz-row-line {
    grid-template-columns: minmax(0, 1fr) auto;
    grid-template-areas: 'name amount';
  }

  .vz-table--month-detail .vz-c-name   { grid-area: name; white-space: normal; }
  .vz-table--month-detail .vz-c-money  { grid-area: value; }
  .vz-table--month-detail .vz-c-qty    { grid-area: slots; }
  .vz-table--month-detail .vz-c-draws  { grid-area: draws; }
  .vz-table--month-detail .vz-c-rate   { grid-area: rate; }
  .vz-table--month-detail .vz-c-amount { grid-area: amount; }

  /* Each figure on a group card is a line: its noun at the card's left edge, the figure at its right
     — `--collect`'s labelled money line, for the same reason. */
  .vz-table--month-detail tbody tr:not(.vz-row-line) td:not(.vz-c-name) {
    justify-self: stretch; /* the noun at the card edge: `.vz-c-money` alone ends its own box */
    display: flex;
    align-items: baseline;
    justify-content: space-between;
    gap: var(--vz-space-3);
    text-align: right;
  }

  .vz-table--month-detail .vz-cell-label {
    align-self: center;
    min-width: 0;
    overflow-wrap: anywhere;
  }

  /* A line row reads at the size of its figure, on one baseline with it; only a GROUP is a heading. */
  .vz-table--month-detail tr.vz-row-line { align-items: baseline; }
  .vz-table--month-detail tr.vz-row-line .vz-c-name { font-size: inherit; font-weight: inherit; }
  .vz-table--month-detail tr.vz-row-line .vz-c-amount { justify-self: end; }
}
