test: cover CaseStudies' two empty-corpus arms with a variant build

[hivecommons-hive]

Test Improvement

src/components/CaseStudies/index.js branches on two document-level fields of
data/case-studies.json that the shipped corpus never produces, so neither arm
runs in a browser:

46   if (!data.generatedAt) return null;          // SyncStatus renders no provenance line
64   const caseStudies = data.caseStudies || [];  // the list the page sorts/filters/counts/maps

The e2e report on main names both: src/components/CaseStudies/index.js | 100.00 | 96.55 | | 46 64 (artifact e2e-coverage id 11383321728, job
End-to-end coverage, main at 7ab301e). The unit suite has them — this is
a browser-coverage gap, not an untested path.

tests/e2e/fixtures/data-variants/case-studies.json (new) — clears
generatedAt and caseStudies. It goes in the variant directory rather
than beside the existing additive overlay at
tests/e2e/fixtures/data/case-studies.json: both fields are document-level on
the one file the one /resources/case-studies page reads, so clearing them in
the ordinary coverage build would swap which arm is covered, trading the
provenance line and all 215 rows — including the two undated studies that
additive overlay exists to render — for these two. The variant site is served
beside the real one, so one Playwright run reaches both and neither displaces
the other. Same shape, and same remedy, as RadarReports in #1098.

tests/e2e/case-studies-variant.spec.js (new) — two cases. The real route
is asserted alongside the variant one on purpose: on its own, "the variant page
has no provenance line and no rows" passes just as well when the page failed to
build, so the pair is what makes it evidence the arms switched. Counts are
read from the corpus through loadSiteData rather than hard-coded, so an edited
overlay fails here instead of leaving a test that asserts nothing.

Verification

• npm run test:unit — 1966 passed, 0 failed at 7ab301e + this diff.
  That includes tests/e2e-data-fixtures.test.mjs's "every committed overlay
  still applies to its data file", which picks the new overlay up with no edit
  to its hardcoded cases lists.

• npm run build:e2e:coverage — succeeds, and the two builds render opposite
  arms:

  	build/resources/case-studies/	build/e2e-coverage-variant/resources/case-studies/
  Mirrored from provenance line	1	0
  <th scope=row> rows	215	0
  results bar	Showing 215 of 215	Showing 0 of 0
  No case studies match those filters.	absent	present
  <option> in the three selects	154	3 (one All … each)
  <th scope=col>	6	6

  Every locator the spec uses was checked against that markup.

• loadSiteData('case-studies.json', …) returns generatedAt: 2026-09-24…
  with 215 studies for the coverage build and generatedAt: null,
  caseStudies: null for the variant build.

• npx prettier --check on both new files — clean.

• The Playwright spec itself could not be executed here: the sandbox has no
  root, so npx playwright install-deps chromium cannot run and
  chrome-headless-shell aborts with
  error while loading shared libraries: libglib-2.0.so.0. CI's
  End-to-end coverage job is the first real execution — the same limitation
  test: let data overlays patch a root-level JSON array with setWhere #1110 records.

What this does not do

It does not move lines 46 and 64 off the uncovered-region list.
getRegionCoverage in tests/tools/e2e-coverage-report.mjs keys a region on
its exact original line:column:endLine:endColumn, and the two builds'
generated code maps back to different coordinates, so the union keeps the real
build's zeros. That was measured on this identical shape in #1097 and is
tracked as #1066 / #1079; once the union folds the coordinates, these arms fold
with no change here. The spec is therefore stated as behaviour rather than as a
coverage claim, and the region gate (--check-source-regions 80) is unaffected.

Coordination

Touches only the two new files above. It does not edit
tests/e2e-data-fixtures.test.mjs (#1110, #1113),
tests/e2e/data-fixtures.spec.js / tests/tools/e2e-data-fixtures.cjs /
tests/e2e/fixtures/data/architectures/catalog.json (#1110),
tests/e2e/fixtures/data-variants/metrics.json (#1117),
tests/e2e/fixtures/data-variants/community-groups.json (#1113),
tests/e2e/fixtures/data-variants/members.json (#1104),
tests/e2e-coverage-report.test.mjs (#1114),
tests/e2e-coverage-scripts.test.mjs (#1128), or any workflow file.
src/components/CaseStudies/index.js is claimed by no open PR.

Related Issue

Closes #1132

---

Filed by quality agent (hold-gated mode). Human review required.

— hive: agent=quality backend=copilot model=claude-opus-5 copilot=1.0.88

[hivecommons-hive]

Important

Held for human review by the hive's ACMM level gate.

This PR was opened by the "quality" agent while Hive policy required a human checkpoint for that agent. Non-outreach agents are held at ACMM L3–L5; the outreach agent is always held because it publishes project-facing communication.

Hive will keep the hold label until a human removes it. Operators can make a deliberate one-off release during an ACMM level change with release_level_holds=true, but level changes never release this hold automatically.