test(evidence): distinguish coverage declarations from execution
Merged
TH
14d ago+76+76+01234567890123456789-71-71-01234567890123456789
Summary
Name coverage guards as declaration and source-index checks so a skipped test or commented command cannot be mistaken for executed coverage. Preserve every catalog completeness assertion and scope execution claims to the appropriate runner, recorded evidence and scenario completion.
7 test/support files changed. No runtime, device lane or reporting infrastructure changed. iOS already shares Android's live-device evidence owner; tvOS has no live-device lane to extend. Closes #4051; part of #4045.
Validation
Head: 6291af80fd62a3064c39b78e3020253ce69f554c.
- All 52 guard tests passed. Independent Terra findings were corrected.
- Exact-head affected validation passed every selected runnable check, including Node integration, macOS coverage and related tests. Receipt:
.tmp/validation/run-HA63Q4/receipt.json,exactCommit: true. - Isolated tvOS guard: baseline, skipped provider and comment-only command each 7 passed. Removing the macOS replay predicate produced 2 failed /5 passed.
- At this SHA, the existing harness ran built-CLI help successfully and recorded completion; selected completeness passed. An unrun snapshot scenario produced missing-command/scenario results and full completeness rejected. Standard report, step history and external SHA identity were retained. This is host execution evidence, not live-device coverage.
- All disposable checkouts were removed and reviewed sources preserved. Proof and counts.
CI at the pushed head remains authoritative.