test(producer): add mov ProRes distributed fixture

[jrusso1020]

Description

Phase 4 of the distributed rendering plan: test fixtures (see DISTRIBUTED-RENDERING-PLAN.md §11 Phase 4 + §10 test strategy). This is PR 4.5 of the Phase 4 remainder, stacked on PR 4.4 (#847) which added png-sequence format support to the harness.

This PR adds the mov ProRes fixture (tests/distributed/mov-prores/). The composition is a 2-second (60-frame) scene at 30fps with a transparent background, a crossfade transition straddling the frame-30 chunk seam, and a continuously rotating SVG icon — same shape as the mp4-h264-sdr and png-sequence fixtures. renderConfig.format: "mov" + chunkSize: 15 produces N=4 chunks, each encoded as ProRes 4444 with alpha (yuva444p10le) and assembled via -c copy.

ProRes is intra-only — every frame is a keyframe — so concat-copy at the frame level is structurally trivial. What this fixture pins is the orthogonal contract: that ffmpeg's concat -c copy across four mov files preserves the QuickTime atoms (moov, mvhd, trak, mdia) intact so the assembled file plays in mov-native editors (Premiere, FCPX, DaVinci) without re-encoding. Both modes produced byte-identical 60-frame output against the in-process baseline (all 100 PSNR checkpoints came back ∞ for both modes).

LFS:

• .gitattributes adds packages/producer/tests/distributed/*/output/output.mov and output.webm patterns. The 3.8 MB ProRes baseline is the main motivation; .webm is included pre-emptively so when webm becomes distributed-supported the path is already configured.

Testing

• bun run --cwd packages/producer docker:test:update mov-prores — baseline rendered inside Dockerfile.test (60 ProRes 4444 frames, 640×360, alpha, ~3.8 MB)
• bun run --cwd packages/producer docker:test mov-prores — in-process passes (100/100 checkpoints byte-identical against baseline)
• bun run --cwd packages/producer docker:test:distributed mov-prores — distributed-simulated passes (100/100 checkpoints byte-identical against baseline)
• bun run --cwd packages/producer docker:test:distributed font-variant-numeric many-cuts gsap-letters-render-compat style-1-prod sub-composition-video mp4-h264-sdr png-sequence mov-prores -- --sequential — full smoke set + all three stacked fixtures pass (8/8)
• bunx oxlint + bunx oxfmt --check clean on changed files
• bunx tsc --noEmit (producer package) clean

🤖 Generated with Claude Code

[jrusso1020]

• test(producer): add chunk-boundary fixtures per first-party adapter #852
• test(producer): add mp4 H.265 SDR distributed fixture #851
• feat(producer): add codec knob to DistributedRenderConfig #850
• test(producer): add mov ProRes distributed fixture #848 👈 (View in Graphite)
• test(producer): add png-sequence distributed fixture #847
• test(producer): add mp4 H.264 SDR distributed fixture #845
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[miguel-heygen]

Re-reviewed as part of the fixture stack. LGTM.

[vanceingalls]

(Intended verdict: APPROVE — posting as COMMENT because the self-stamp guard blocked the approval action; treat as a green review modulo the notes below.)

Clean addition. Matches the established #845/#847 stack pattern; correctly identifies that ProRes is intra-only and pins the orthogonal -c copy + QuickTime-atom contract, not state-continuity. Validated end-to-end (in-process + distributed-simulated both byte-identical against the baseline).

Strengths

• meta.json:11-15 — chunkSize: 15 correctly produces N=4 chunks at 60 frames, and the GSAP crossfade at src/index.html:91-92 (0.9s → 1.1s) straddles the frame-30 seam by design, so the fixture meaningfully exercises chunk boundaries even though ProRes is keyframe-only.
• Fixture leans entirely on existing wiring from #847 (harness outputFormat branching + runDistributedSimulatedRender format cast at regression-harness.ts:805) — no plumbing duplication.
• Description is unusually precise about what this fixture pins vs. what it doesn't (atom preservation through concat-copy, not frame inter-dependency). That framing prevents the next contributor from re-litigating "is this redundant with mp4-h264-sdr?"

Findings

important

• .gitattributes:10 — the output.webm LFS pattern is added pre-emptively, but webm is actively rejected by checkDistributedSupport (regression-harness-distributed.ts) and plan() throws FORMAT_NOT_SUPPORTED_IN_DISTRIBUTED (distributed/plan.ts:270). There is no current or near-future code path that produces a tests/distributed/*/output/output.webm. Adding LFS rules for a path that can't exist is scope creep — recommend deferring to whichever PR actually enables webm in distributed mode. Failure mode is minor (an inert LFS pattern), but it muddies the audit trail when the webm fixture eventually does land and someone wonders "did this PR add the LFS rule, or is it the same one?".
• meta.json:3 description — worth one sentence noting that the PSNR comparator (psnrAtCheckpoint in regression-harness.ts:415) does not include the alpha plane in average: by default (ffmpeg's psnr filter measures Y/U/V only). For this fixture byte-equality + maxFrameFailures: 0 catches drift, but a future author cribbing this fixture and authoring minPsnr: 30 against an alpha-only regression would get a false pass. Forward-looking note prevents that footgun.

nits

• src/index.html is missing two comments siblings have: the "no <audio> element on purpose — AAC frame quantization extends container duration past 2.0s and trips the last PSNR checkpoint" rationale (mp4-h264-sdr/src/index.html:92-98), and the explicit annotation that frames {15, 30, 45} sit inside the crossfade window (mp4-h264-sdr/src/index.html:103-105). Cribbing those over keeps the three fixtures' surface explanations parallel.
• .label text is PHASE here but CHUNK in mp4-h264-sdr. Cosmetic, but stack coherence is cheap.
• minAudioCorrelation / maxAudioLagWindows in meta.json:7-8 are dead config for silent compositions — the audio block short-circuits when extracted PCM is empty. Carried over from the sibling pattern; not actionable here, just calling it out for whoever revisits the meta schema.

Verdict: APPROVE (posted as COMMENT — see top note)
Reasoning: Fixture is correct, well-scoped, and lands cleanly on the harness wiring #847 set up. The webm LFS rule and alpha-PSNR docstring are worth addressing but neither blocks merge — both are easier to fix in a follow-up than to gate a fixture PR on.

Review by Vai

[jrusso1020]

Thanks @vanceingalls — addressed comment-parity in commit 8628a683. The mov-prores fixture now carries the same chunk-seam mapping + ProRes-intra-only rationale comments as the mp4 siblings.

On the output.webm LFS pattern — kept it intact. The team's stance is that any video file in the repo MUST go through LFS, so the pattern is a safety net to prevent a future webm fixture from accidentally landing as a regular binary. Cost of leaving it is zero (inert pattern); cost of removing it is one round-trip when webm distributed support lands.

The alpha-plane-PSNR docstring nit is a fair forward-looking note but doesn't load-bear here (this fixture uses byte equality, not PSNR) — leaving for the v1.5 PR that enables alpha-aware comparison.

The base branch was changed.