test(app-shell,scripts): @object-ui/i18n joins COVERED_SPECIFIERS (objectui#7337) - #7920

Fixes #7337

The ruled follow-up (option A) to PR #7889: the @object-ui/i18n slice of the
check-vi-mock-inherit ratchet is finished, so the gate — not a pin file — now
holds that specifier's frozen population at zero.

Three steps, one PR, in the only order that works. Converting first and flipping
second is not stylistic: with a frozen factory still in the tree the flip fails
check:vi-mock-inherit on the very next run.

1. packages/app-shell/src/views/__tests__/DeclaredActionsBar.test.tsx — the one
   remaining frozen vi.mock('@object-ui/i18n', ...) factory obtains the real
   module and spreads it, keeping the three overrides the suite needs
   (useObjectLabel, useObjectTranslation, pickLocalized). Same spelling as
   its already-converted sibling DeclaredActionsBar.overrideAffordance.test.tsx.
2. scripts/__tests__/check-vi-mock-inherit.test.ts — the case that recorded WHY
   the specifier was not covered yet is deleted, along with the two expectations
   that encoded the un-flipped state (the held-file carve-out, and a census
   expectation that allowed one non-inheriting site).
3. scripts/check-vi-mock-inherit.mjs — @object-ui/i18n joins
   COVERED_SPECIFIERS, and the header stops saying the covered set has exactly
   one member.

Measured, before and after, on this branch

The one frozen site before was DeclaredActionsBar.test.tsx:65, exactly as the
release comment predicted. Zero sites moved the other way: the frozen sets
were diffed per site (file plus mocked specifier, line deliberately omitted
because an edit above a site shifts its line without moving its verdict), and the
only difference is that one site leaving.

Note for whoever re-derives these: the all-specifier population here is taken over
all 39 @object-ui/* workspace packages, read out of packages/*/package.json.
The last reading on the merged head used a 21-specifier set and reported 312
frozen / 627 covered. The delta is entirely definitional — inherits agrees at
315 in both — so the extra 3 are frozen sites on specifiers the narrower set never
named, not a change in the tree.

Gate verdict line after the flip:

check-vi-mock-inherit: OK (4331 tracked source file(s), 2610 test-named; 586 carry
a mock; 223 call site(s) on @object-ui/react, @object-ui/i18n judged (223 inherit,
0 auto-mocked); 409 other workspace, 229 external, 844 local, 0 non-static, 3
embedded in a string literal -- all out of scope).

223 = 131 on @object-ui/react plus 92 on @object-ui/i18n.

Reverse verification, two legs

Both legs: commit first, mutate, prove the bytes reached disk by counting the
injected and the deleted text, run, restore with git checkout HEAD -- PATH,
prove the restore by an empty git diff HEAD and a blob hash equal to HEAD's.
Restore runs from a trap on EXIT/INT/TERM with absolute paths.

Leg A — re-freeze the factory, keep the flip. The gate goes RED, exit 1,
naming the site:

- packages/app-shell/src/views/__tests__/DeclaredActionsBar.test.tsx:65 -- vi.mock("@object-ui/i18n")
  the factory never obtains the real module

Restored blob 67cdf539b, git diff HEAD empty.

Leg B — drop the specifier from the covered set, keep the conversion. This is
the non-vacuity check for the one case this PR ADDS, and it measures the exact
failure mode that case exists for: the gate stays GREEN (exit 0, 131 judged
instead of 223) because every @object-ui/i18n call site becomes unjudged — a
ratchet retiring in silence. The pin file is what reddens, with exactly one
failing case, the new one:

FAIL scripts/__tests__/check-vi-mock-inherit.test.ts > the sweep ... > the specifier IS in COVERED_SPECIFIERS
AssertionError: the sweep landed; removing the specifier retires the ratchet silently:
expected [ '@object-ui/react' ] to include '@object-ui/i18n'
Tests  1 failed | 68 passed (69)

Restored blob 1cea0ddf4, git diff HEAD empty.

One case changed beyond the three steps, and why

puts the census in the verdict built its expected string from a file-local
constant naming a single specifier, while the gate's verdict line joins the whole
covered set. It read as an assertion about the verdict line and was really an
assertion about the set's SIZE — measured: it was the only failure on the first
run after the flip. It now derives the names from COVERED_SPECIFIERS itself, so
the next widening does not pay for it. The rest of the death pin (legs 1/2/3, the
non-vacuity control, the replay case and the JSON-reporter verdicts) is untouched:
69 cases before, 69 after.

Gates, each with its own printed verdict

Exit codes captured by redirect-then-capture, never through a pipe.

Every heavy run went through scripts/pm/os-verify-lock.sh; verdicts read from its
VERDICT command-exit line, never a bare shell status. The dependency closure was
built first (pnpm --filter '@object-ui/app-shell^...' build) — without it
type-check fails on missing sibling declaration files, which reads exactly like
a break this diff caused.

Union re-run at the final commit 4c871b0e3, working tree clean.

Scope

Four files: the converted test, the pin file, the gate, and one
empty-frontmatter changeset. No other specifier is swept — objectui#6892 stays
serial behind this card. No exemption table and no per-site allowlist; the pin
that forbids one still passes.

Part of the worklist tracked in objectui#6892. Related: PR #7889, PR #7846.

Generated by Claude Code