# PR #22788 evidence manifest (pass 2, fs-08, 2026-10-22)
B = 8d3b061121fe65669d5d43d011b48cd3027a4108 + fixture overlay (status: M acp-mock-agent.ts, A timeline-follow-fixture.ts). A = 6f2d6c63e3f252651c0221c8a3dcf9d9f950dbbd.
Fixture print sha256 B = A = c26f312ce0ea3f4c3fd33bf5e9b431de178535a49fe93318a9cdd51d35f35b87. Hold script sha256 0f3b614fce0e30de95131eac0abbae07b44c30efb4d6a2b8a8a45522c3d1c6fb. Driver commit 0af183bec7 (apps/server/scripts/timeline-follow-evidence.ts).
Environment: Linux 6.1.102 x86_64, Node 24.22.1, Playwright 1.60.0, chromium headless shell 148.0.7778.96, 1440x900 DPR1 (W rows, ST1/ST2), en-US/UTC, page.clock fixed 2026-01-01T12:00Z. Hold ~2.6–2.9 s, 143–178 RAF frames. Servers: B 13814/5774, A 13815/5775, isolated homes /tmp/ev-22788/run/home-{B,A}.

Hydration: TEST-DRIVER hold, labelled, no source change, identical on B and A. Deviation from the matrix: the matrix's WS-only hold engaged but never produced the loading placeholder, because this client loads the thread via HTTP GET /api/orchestration/threads/<id>/bounded before subscribeThread. The driver therefore also holds that route with page.route. The matrix's hold.js also re-dispatched held frames to the first socket (Vite HMR) instead of the receiving one, which killed the stream; fixed in the driver.

| Row | B | A | Verdict |
|---|---|---|---|
| S0 transcript | S0-B.txt | S0-A.txt | PASS (sha match, seed exit 0, waiting.json entry→…→resumed, .done after each release) |
| W1 PageUp | W1-B.mp4 + narration | W1-A.mp4 + narration | B snap NOT REPRODUCED; A hold: listeners 4/1 vs B 0/0, but A's hold criterion not proven (see notes) |
| W1p pill | W1p-B (no pill on B) | W1p-A | A pill resume: gap 16 then 160 at S_res → FAIL (gap>2) |
| W2 wheel | W2-B | W2-A | B snap NOT REPRODUCED; A listeners 4/1, handleWheel on timeline |
| W3 no-input control | W3-B | W3-A | FAIL on both: gap 0→1117→1229 with no input, so live follow did not track growth within 500 ms on either revision |
| Phase-range indexing | | | FAIL: overlay color matched within 0.2 s for only some marks (see phase-range-verification.json); W1 clips 0/12, likely the recordVideo start offset |
| ST1 1440 light | C-mounted/C-held B,A + Diff0/DiffT | | UNVERIFIED: C-mounted deltas not attributed (DiffT 46591 px; sidebar thread count and labels differ) |
| ST2 1440 dark | same | | UNVERIFIED: C-mounted DiffT 6390 px, sidebar only (bbox x≤479) |
| ST3/ST4 390 | not captured | | UNVERIFIED: out of time; at 390 the companion was off-screen or in a collapsed sidebar |
Overall: INSUFFICIENT.