chore(frontend): lint:ceiling fails on a clean master — stale no-irregular-whitespace count

[alex-dembele]

Problem

npm run lint:ceiling fails on master, with no change in the working tree:

$ npm run lint:ceiling
ESLint errors: 318 (ceiling 321)

✗ ESLint regressed. These rules got worse:

    no-irregular-whitespace: 3 (ceiling 1, +2)

The total is comfortably under the ceiling. The failure is entirely a stale per-rule
entry: .lint-ceiling.json records "no-irregular-whitespace": 1, and the tree has three.

The three are all pre-existing, and none of them is a defect — they are deliberate non-breaking
spaces in test fixtures and expectations:

• frontend/e2e/empty-states.spec.ts:167
• frontend/src/i18n/__tests__/format.test.ts:17
• frontend/src/i18n/__tests__/translate.test.ts:19

The last two are the i18n formatter's own tests, which assert French number formatting — and
French number formatting is a narrow no-break space. Removing the character would make the
test assert the wrong thing.

Found while working #438: the gate flagged a "regression" on a branch that had actually
reduced total ESLint errors from 318 to 315.

Why this matters beyond the noise

D-022 set this ratchet deliberately: "the frontend lint gate is ratcheted at 321, not swept and
not downgraded." A ratchet that fails on an untouched tree trains everyone to ignore it, which
is precisely how the next real regression gets waved through.

Acceptance criteria

1. Given master with a clean working tree, When npm run lint:ceiling runs, Then it passes.
2. Given the three sites above, When they are read, Then either the non-breaking space is kept
   and the rule is configured to allow it in test fixtures (skipStrings, or an override for
   **/__tests__/** and e2e/**), or each site carries a targeted disable comment saying why
   the character is intentional. The character must not be silently replaced with a normal
   space in the i18n tests: that would make them assert formatting French does not use.
3. Given .lint-ceiling.json, When it is updated, Then the total is not raised — only the
   stale per-rule count is corrected to match a tree that genuinely has no new violations.
4. Given a deliberately introduced new violation of any ratcheted rule, When the gate runs,
   Then it still fails. The fix must not disarm the ratchet.

Definition of Done

• npm run lint:ceiling green on a clean master
• Criterion 4 demonstrated in the PR (paste the failing output for a deliberate violation,
  then its removal)
• No change to the total ceiling of 321
• The i18n tests still assert French formatting with its real characters

Out of scope

• Reducing the 315 existing errors. That is the ratchet's job over time, not this issue's.
• Raising the ceiling. D-022 is explicit.

[alex-dembele]

claude-opus-5 — 2026-09-11

Not picking this up, but the two offenders are located, which should make it a five-minute fix.

src/i18n/__tests__/format.test.ts:17:41     Irregular whitespace not allowed.
src/i18n/__tests__/translate.test.ts:19:41  Irregular whitespace not allowed.

Both are inside i18n tests, at column 41 — almost certainly intentional non-breaking
spaces in French formatting assertions (Intl.NumberFormat('fr') emits U+202F / U+00A0 as the
group separator). If so, deleting the character would break the assertion it exists to make,
and the right fix is a scoped eslint-disable-next-line no-irregular-whitespace with a comment
saying why the character is deliberate — not a change to the string.

Worth confirming that reading before touching them; I have not.

Context: hit while purging dead code on #637. That PR drops ESLint from 312 errors to 184
and re-freezes .lint-ceiling.json at the new floor, but leaves no-irregular-whitespace at
its existing ceiling of 1 rather than the measured 3. npm run lint:ceiling -- --update
raises it to 3, which is what this script's own failure message forbids ("Fix them — do not
raise the ceiling"), so I put that one entry back by hand. The gate therefore still fails on
this issue and only this issue — which is the honest state until it is fixed.

[alex-dembele]

claude — 2026-09-17

Done — nothing changed in code. This comment widens the evidence: on master 53c5f29, lint:ceiling now fails on three rules, not one. CI shows the same result (run 35206997940, job "Frontend Lint & Format").

$ npm run lint:ceiling
ESLint errors: 185 (ceiling 185)
✗ ESLint regressed. These rules got worse:
    no-irregular-whitespace: 3 (ceiling 1, +2)
    react-hooks/set-state-in-effect: 17 (ceiling 16, +1)
    react-hooks/refs: 2 (ceiling 1, +1)

Verified — npx eslint src -f json on 53c5f29, filtered to the three rules:

• no-irregular-whitespace: src/i18n/__tests__/format.test.ts:17, src/i18n/__tests__/translate.test.ts:19 (as already located above)
• react-hooks/refs (2): src/shared/ds/TagInput.tsx:151, src/shared/useHotkeys.ts:29. TagInput arrived in 01b4ba9 (design(ds): TagInput does not exist — and four other components #438 says to reuse #631) on 2026-09-11, after the ceiling was last written (4912eb2, same day), so it is the likely +1. Not proven.
• react-hooks/set-state-in-effect (17): App.tsx:359, CommandPalette.tsx:100,112, MFAPostAhaPrompt.tsx:64, dashboard/shared.tsx:36, PostMortemPanel.tsx:54, ProductTour.tsx:120, wizard/steps.tsx:85,279,581, BoardReportPage.tsx:68,306, ReportConfigurator.tsx:68, RiskWeightsSettings.tsx:79, MFAPolicyPanel.tsx:43, SettingsScreen.tsx:289, shared/ui.tsx:25. Which one is the +1 is not established; files touched since the ceiling was written: App.tsx (bc6b428), PostMortemPanel.tsx (c6231bb), wizard/steps.tsx (75f1467), ReportConfigurator.tsx (04f6e04).

Criteria — unchanged. Criterion 1 ("passes on a clean master") now needs all three rules brought back under the ceiling, not only the whitespace count.

Next — fix the two whitespace hits as described above. Then use git log -S / blame on the lines listed to find the new refs and set-state-in-effect hits, and fix them rather than raising the ceiling.

Blocked on — nothing. Found while verifying #524 (PR #713); that PR neither adds nor removes any of these.