Service helper test pruning — VUH-1925

Reviewed unit-only cut on main base 1d8ae045f89141536ff3e57247a0ce660422ae34: 91 → 67 test lines; four → three cases. Product code, exports, fixtures and E2E/integration/golden files are unchanged. See the per-file keep/cut review.

The first loses a wording-only case. The second loses an exact text assertion while retaining its blank-event case and bounded overlay case. Runtime coverage in play-voice.test.ts retains the meaningful reporting fields, room-owned speech, journal provenance and lossless experience boundary.

The baseline root check:landing used its own native --changed selection. Temporary comments selected both unchanged test files, then were removed before the cut. Exit 0, sourceStable true, four passing cases in two files, 65.943s. One uncached service compiler took 24.410s. Wrapper admission waited for fleet capacity; that queue wait is outside the gate's reported elapsed time.

Moss resumed on main 4126ed0abe75fde868ec757f444b6224a2f6995c. Both original test files and the retained play-voice.test.ts runtime cases are byte-identical to the baseline; the relevant model extension and room-event helpers are unchanged. The before gate and 13/13 retained runtime proof are reused. Main's new dependency required a frozen-lockfile install before the landing gate. The different base and shared-machine/compiler-cache conditions prevent a matched speedup claim.

Landed on main as 3ef50411dd49bf34e86a8eb0fb85ab169d3006dc, cherry-picked onto fetched main 6f7e23aa2139695a5abc45a2a5a632304b6c2a4f after VUH-1948. The fresh-checkout narrow run passed 16/16 (both touched files plus all 13 play-voice.test.ts runtime cases). The committed-head root gate passed 3/3 in both touched files, exit 0, sourceStable true, 131.711 seconds. All phases passed; Vox lint compiled fresh dependencies (85.156s phase) and the service typecheck was uncached (21.280s phase). Final pull left HEAD/base unchanged before push.

Moss's old-base after gate also completed: 3/3, exit 0, sourceStable true, 36.525s on d4476f1e against 4126ed0a. It is preserved as earlier evidence; landing acceptance uses the fresh-checkout gate. A premature narrow request was canceled while queued before execution because dependencies were absent; its SIGTERM/exit 143 trace is retained. The frozen install passed in 9.7s.

Final committed-head root gate and retained runtime results are recorded on VUH-1925, together with the raw evidence archive. Local receipts: .local/vuh-1925/service-before-gate.json, service-before-vitest.json, service-after-gate.json, service-after-vitest.json, service-retained-vitest.json and service-counts.json. These shared-machine before/after values are single observations, not a claimed repeatable speedup. The gate above did not reuse the service compiler cache.

Service conventional test source on the measurement base: 510 files, 157,294 → 157,270 lines. Rust inline production lines are not part of this count. On resumed main: 511 files, 158,238 → 158,214 lines. The same two-file 24-line cut accounts for the change; upstream additions are not pruning savings. At final landing: 512 service test files, 158,478 → 158,454 lines.

Running unit-pruning subtotal after this batch: five batches, 659 test lines and 38 cases removed. The schema fixture separately lost 50,178 lines with assertions preserved. The inventory is incomplete; this is a batch result, not closure.

Raw before/after gates, narrow/runtime checks, logs, counts and the committed diff are archived in the object listed in evidence.json. Raw bytes stay outside git.

Archive: clankie://evidence/sha256/1b8ff3ec7ed3baeb6ad185c9c23457c8e04b6adc24058f743bbac8b13986c076. Upload receipt is applied; manifest records its hash and size.