# Independent reuse review — brief-04-attempt-01

Date: September 5, 2026. Read-only packet review; evidence is written only here.

## Recommendation

**Qualified success for independent, documented recipe reuse; not an unqualified source-fidelity or strict write-isolation pass.** The consumer selected a suitable treatment through the catalog, substituted the exact brief copy and all six supplied assets, and produced a correctly bound silent movie without changing the delivered implementation. The native movement, in-progress entry, masked row coexistence, stationary suffix, hard closing-line switch, and short ending survive the substitution in the inspected evidence. I found no consumer-specific copy, asset-order, clipping, retiming, or implementation-edit defect.

The stronger claim that this trial establishes fully source-matched reuse remains **not established**: inherited pill geometry differs from the native source, exact font/fade reconstruction remains qualified, and early consumer npm runtime writes outside the packet cannot be certified absent. These are separate questions, not reasons to erase the successful reuse evidence or to infer undocumented help.

This is a recommendation, **not acceptance on the user's behalf**. No draft inherits approval. No completed-example credit, publication credit, catalog/status change, or implementation correction is made.

## Scope and integrity

- Packet: `benchmarks/remotion-library/trials/packets/brief-04-attempt-01`. Its `packet.json` SHA-256 is exactly `ae5cdee047562b94ac6cc7a3aa8447fa27e7475d01b21cdb2a0296a35c499cf9`.
- Read only the delivered packet's relevant brief, guide/API, selected catalog/recipe, actual submission, selected source evidence, and authorized main audit files, plus root/user instructions. No author handoffs, external research links, surrounding source research, other trials, or original builder context were consulted. Other draft renders/findings did not supply a verdict.
- All **22,128 packet file/symlink entries**, including installed dependencies, caches and consumer outputs, were inventoried before and after. Regular files were SHA-256 hashed and symlink targets recorded. **Zero added, removed, modified, or mtime-only-changed packet entries.** All **7,686 frozen protected files** also match the delivered packet manifest.
- `inputs-before.json`, `inputs-after.json` and `input-comparison.json` retain that check. Additional authorized audit logs/browser executable use first-recorded baselines in `inputs-audit-arrival.json` and `inputs-tool-baseline.json`; all compared inputs are unchanged. Tool entry-point hashes are not a full toolchain attestation.
- Main audit: `mechanicalPass=true`, `trialPassed=false`, `immutableChanges=[]`; its watcher summary reports zero protected events. I read that summary, not the out-of-scope raw watcher. Earlier operator audit logs both show an invalid project-relative output-path failure; the corrected project-relative audit succeeds. Those are operator invocation failures, not consumer render failures.

## Independent mechanical and content checks

Ran the delivered verifier myself from the packet's `project/`:

```text
npm --silent run verify -- --out out/trial/nook.mp4
```

Reviewer HOME, npm/XDG caches and temporary directories were configured inside this evidence directory. `verification.json` records exit success and `matches-current-code-input-assets`; fidelity remains explicitly `not-assessed`. The verifier used the **actual sibling** `out/trial/nook.manifest.json`, not a substitute manifest. Its embedded recipe equals the delivered recipe; embedded content equals the editable JSON. Frozen hashes also cover shared code, config, tracks, catalog and recipe bytes.

| Artifact | Independently checked result |
|---|---|
| MP4 | 1920×1080, H.264/yuv420p; 300 frames; 60000/1001 fps; 5.005 s; no audio |
| MP4 SHA-256 | `cfc96ffc93656b6ae475a8058c48132575928e707d6e9490e11142b91a4a8ce4` |
| Sibling manifest SHA-256 | `145ca0b076ea0bd02c9eae57f6fa2655b32d985e6a97e83ef6be0d7292c7e68c` |
| Copy | `Form`, `Craft`, `Print`, `Build`, `Type`, `Mark`; `by Nook`; `Built with care` |
| Assets | All six originals match their supplied and copied SHA-256s in the correct slots; rights statements retained |
| Editorial additions | Zero before/after holds; no narration, introduction, extra dissolve or replacement ending |

The broad catalog query returns five typography drafts; the documented vertical-column/image/≤6-second query returns only `type-masked-name-reel`. Direct Content JSON is the documented route (`contentBuilder=null`). Only the fourteen icon/text fields differ from defaults; `accent` and `fade` remain unchanged. The retained commands and disclosure support independent selection, not a prescribed recipe ID.

Supplementary loaded-Geist measurements reproduce fit at unchanged sizes: rows 64 px, suffix 100 px, closing 98 px. Row advances are 139.30–151.52 px against 181–223 px boxes; suffix 389.53/610 px; closing 662.20/960 px. No missing letters, wrapping or clipping appeared in inspected frames. These are fit measurements, **not audience-reading or accessibility certification**. Form remains a faint upper neighbor at entry rather than receiving a centered beat, as the brief explicitly allows. The short inherited reading windows and 0.85085-second closing remain limitations for a real audience.

## Moving and native-frame inspection

The review's first two tool calls were **T3 status, then open**; both explicitly reported no host. I used the authorized external Playwright entry and `[LOCAL_PATH_REDACTED]` solely for reviewer inspection. No packages were installed, builder advice obtained, or toolkit code altered. `reviewer-preflight.json` records startup/helper failures and recovery, including the isolated-HOME and long-socket-path workarounds.

The moving comparator is the delivered 960×540 `evidence/type-masked-name-reel/review02/source.mp4`, SHA-256 `458b420e6ab756871aa0bdd17fa65970614c58310226fdf26e22e0c71d0742e4`. It maps to Apps in ChatGPT native frames **[3596,3896)**. I verified all 100 delivered native-source PNG hashes against `comparison.json`, and used those **1920×1080 native PNGs**, not enlarged excerpt frames, for boundary/intermediate comparisons and source pixel measurements. The full-film hash is declared provenance, not independently reverified here.

Both movies were played separately from start to `ended`, once at 1× and once at 0.25×, with frame callbacks, media events, wall/media times and screenshot brackets retained:

| Isolated run | Wall seconds | Frame callbacks | Browser-reported dropped / total | Corrupted |
|---|---:|---:|---:|---:|
| Source 1× | 5.0491 | 279 | 0 / 300 | 0 |
| Output 1× | 5.0449 | 257 | 0 / 300 | 0 |
| Source 0.25× | 20.0903 | 299 | 0 / 300 | 0 |
| Output 0.25× | 20.1175 | 299 | 0 / 300 | 0 |

Each run reached 5.005 s. These counters do not prove every frame was observed. This is **machine playback plus visual inspection of advancing screenshots and exact-frame evidence**, not continuous human viewing. All 44 captured playback screenshots were visually inspected through four reduced grids; screenshot media times are before/after brackets, never asserted exact frame addresses. The consumer's earlier anomalous paired-source drop counters remain a limitation of its own evidence; these independent isolated runs do not rewrite them.

I decoded all 300 frames of each movie, but visually visited **48 exact native-source/output pairs**, not all 300:

```text
0,1,2,15,35,37; 43,45,59,60,85,87; 96,97,98,99,100,118;
136,138,143,145,164,183; 185,200,203,214,235,236;
238,240,242,244,246,247; 248,249,250,252,254,256;
278,280,282,297,298,299.
```

Source frame = local frame + 3596; local seconds = frame × 1001/60000. The eight paired grids are reduced to 640×360 per image. Native source and output frames **0,98,248,249,299** were additionally opened individually at their original 1920×1080 dimensions. `visited-frames.json`, `sheet-map.json` and `playback-sheet-map.json` distinguish exact frames, callback observations, and screenshot brackets.

## Findings

1. **Native choreography is retained in the inspected sequence.** Entry is already moving; labels/icons/pills travel together beneath the fixed fade, with old and incoming rows coexisting rather than typewriter replacement. The fast 97→98 travel exists in both native source and output. Suffix placement stays separate from row movement until the inherited departure. At local **248/source 3844** the reel remains; at **249/source 3845 (4.15415 s)** the complete closing line replaces it with no dissolve or partial-word beat. Ending occupies **[249,300)**: 51 frames, 0.85085 s. Both closing lines settle left and expand slightly late; thresholded ink moves left 39 source pixels versus 38 output pixels from 249 to 299. Both encoded timelines have 300 increasing PTS values on the same rational clock, within 0.334 µs of decimal print precision.

2. **Residual source-fit mismatch: pill geometry is not source-identical.** Native y=550 scanlines, grayscale <249, give source/output spans of **436/384 px at local 15**, **404/362 at 59**, **355/394 at 118**, **384/352 at 164**, and **350/352 at 214**. These are measured horizontal visible spans, not recovered vector dimensions. The narrower Craft/Print pills and wider Build pill are visible in the paired grids. This is inherited fixed recipe geometry, **not a consumer geometry edit**. Different label glyph lengths, new icon shapes and the deliberate teal accent are expected substitutions and are not themselves failures. The remaining pill fit prevents calling this a pixel-faithful source reconstruction.

3. **Font and ghost treatment remain qualified.** Geist is an explicitly declared substitute, not the recovered source font. Ghost-row contrast/appearance differs, particularly the additional faint lower neighbor at entry and the exiting upper row around 98. Changed artwork/color also affects that appearance; this review does not isolate an exact alpha-profile error from those differences. The delivered white-gradient overlay is retained; general-purpose alpha-mask fidelity, proprietary easing/shutter and exact source-font fidelity are unproven. The catalog's older claim that final-line onset is still unfitted is too broad for this generation: the native 248/249 boundary matches independently. That does not resolve its other source-fit caveats or promote the draft.

4. **Execution-scope uncertainty is real but separate.** Consumer log 024 records unset npm/XDG overrides and effective default cache `[LOCAL_PATH_REDACTED]` for early commands. Installation and rendering explicitly use packet-local npm caches; later logging pins caches. Possible earlier outside runtime logs/cache writes were not audited and cannot be retroactively certified. **Strict all-writes-local admissibility remains uncertified**, not proven violated. This is not evidence of implementation edits, outside implementation advice, or builder contact. Frozen hashes, the main watch summary, retained commands and consumer disclosure support no implementation edits/no undocumented help, but unsigned logs and hashes cannot prove all historical reads or communications.

## Limits and evidence

No actual preceding/following shot was supplied for this standalone card; only its entry, internal cut and ending were assessed. No target-room/display, audience, accessibility, legal-rights, full-film native decode, or complete browser/toolchain audit was performed. The supplied Morrow-named authorship statements were preserved, not independently certified or rewritten as Nook authorship.

Primary evidence: `report.json`, `verification.json`, `independent-mechanics.json`, `input-comparison.json`, `visited-frames.json`, `pixel-and-timing-measurements.json`, `text-metrics.json`, `playback-summary.json`, `playback/*/playback.json`, `sheets/`, and `evidence-hashes.json`. Detailed commands and environment overrides are in `mechanical-commands.json`, `analysis-commands.json`, `inspect.mjs` and `reviewer-preflight.json`.

**Disposition:** retain this as qualified evidence that an independent consumer can reuse the delivered API and preserve its choreography. Do not turn it into an unqualified source-native approval, a strict cache-isolation certification, or completion/publication credit.