test(docs): hold the roster name column to the view labels the app ships

[os-trump]

Fixes #1326

The measurement first, because it is what the card asked for

55 name cells across the ten English roster pages. ZERO false positives.

Page	name cells	unresolved
sales/accounts.mdx	6	0
sales/contacts.mdx	3	0
sales/leads.mdx	8	0
sales/opportunities.mdx	9	0
sales/activities.mdx	8	0
sales/quotes.mdx	3	0
service/cases.mdx	8	0
marketing/campaigns.mdx	4	0
revenue/contracts.mdx	4	0
revenue/products.mdx	2	0
total	55	0

The guard docstring records 13 false positives, and the card was right that
the number does not transfer: those 13 were raised by the rule at its widest
— every bolded run anywhere in the section. Narrowed to the name column of
the roster table, every one of the 13 falls outside the rule reach, and it is
not luck:

• column and filter names (**Health Score**, **Annual Revenue of $10M or more**) live in the What it shows column, one column over;
• the whole-sentence bolds (**Status is the filter you already have.**) are
  bullets beneath the table, not rows in it;
• service/cases names its six retired non-views in a bullet list below the
  table — the page that was RIGHT is the page this rule never reads;
• the TAB-label shape the docstring lists as something to tolerate is the
  fourth category, and it is not a legitimate exception at all. It was the #760 published to customers: six docs pages print a "Tab" column of strings the console never renders, and one asserts the false model in prose before tabulating it eight times #1318
  defect, in the exact page cited as carrying it.

Emoji-carrying labels (⚠️ At-Risk Accounts, 🔥 Hot Leads, ⏰ SLA at Risk,
📅 My Priority Tasks, ⚠️ Stale Opportunities · Longest in Stage First) all
resolve verbatim — the switcher prints the label as authored, and the pages
already spell it that way. Em-dash labels (Unassigned — triage) likewise.

⇒ Small, and none of it exceptional: an exemption map is not needed at all.
premise_still_valid: true, and the narrow route is viable.

The interaction the PM flagged: measured, and there is none

test/docs-quick-tour-navigation.test.ts reserves bold on
getting-started/quick-tour for names the app really carries, and a name-column
rule requiring bold reads like a head-on collision. It is not one:
quick-tour carries no ## Standard list views heading at all — its
sections are the seven numbered tour steps — so rosterOf returns null for
it, vacuity guard #1 never demands it be mapped, and it is not in PAGE_OBJECT.
The two rules never see the other page. No exemption is needed, and adding
one would document a conflict that does not exist.

What the rule is

Every markdown table body row inside a roster section must open its first
column with a bolded name, and that name must be a label the registered
views of the object actually ship. Three assertions and one vacuity floor:

1. every mapped page must yield at least one name cell (a page that switched to
   a bulleted roster, or a table this parser stopped recognising, would
   otherwise pass by checking nothing);
2. every roster row name column must be bold — de-bolding is not an escape
   hatch;
3. every bolded name must resolve against LABELS, derived from the registered
   stack, not written down here.

The header row is found by the delimiter row below it rather than by its
wording, so revenue/contracts heading its first column Tab and the other
nine heading theirs View both work without either spelling being hard-coded.

Reverse verification — both directions, which is the point of the card

Green-on-main alone is what coverage already achieved while the whole defect
was on disk, so it proves nothing by itself.

HEAD blob:     0f7117ec81c60fbd219fe3e554e379213ca2fc2f
pre-1322 blob: 3fc15685442534c2bace7716571d7a57926b7530
on disk after mutation: 3fc15685442534c2bace7716571d7a57926b7530
MUTATION LANDED: 3fc1568... == pre-1322 3fc1568..., != HEAD 0f7117e...

With content/docs/sales/activities.mdx restored to its pre-PR-#1322 bytes
(794d6fe~1) — hashed on disk before any verdict was read:

 × the name column of every roster names only views the app ships (#1326)
   content/docs/sales/activities.mdx names All, and crm_task ships no view with that label
   ... Board / Schedule / Plan / Worklog / My Tasks / Priority / Backlog
 Test Files  1 failed (1)
      Tests  1 failed | 4 passed (5)

8 red, and the other four tests — coverage included — stayed green,
which is the claim of the card reproduced under the new rule. The rule is
section-scoped: an unrelated form-section table in the same file that also opens
a row with **Schedule** was correctly not read.

Restored with git checkout HEAD -- PATH (never the bare form, which restores
from the index) and proved: git diff HEAD empty, git status --short empty,
git hash-object back to 0f7117e.

No build sits between the mutation and the verdict here — the suite reads the
.mdx off disk at test time and imports objectstack.config.ts through the
vitest transform, so there is no dist/ for a stale artifact to hide in.

What this does not touch

Purely additive: 154 insertions, 0 deletions. The coverage rule, the
structural equal-entry-count rule across faces, and both existing vacuity guards
are untouched. English faces only — the split label column on the translated
faces is open decision card #1329 and is not decided here. No file under
content/docs/ was changed by this PR.

Gates

pnpm verify at 4e02e04, exit 0:

✓ i18n lint gate: 0 `i18n/missing-*` issues
✓ source hygiene clean
 Test Files  142 passed (142)
      Tests  3003 passed | 1 skipped (3004)

Changeset: empty frontmatter — the sanctioned releases-nothing declaration that
.github/workflows/changeset-check.yml documents. Test guard only; no src/
metadata moved.

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
hotcrm	Ignored		Aug 27, 2026 4:23am