Brief 03 attempt 01 — independent reuse review

Trial verdict: PASS within the completed consumer trial's scope (trialPass: true). No newly introduced material adaptation, timing, copy, artwork or continuity defect was found. The supplied consumer output satisfies the brief. This is not user acceptance, a source-identical/completed-example verdict, or publication permission. The inherited paragraph minor residual and the underline source-review limits remain explicit below.

Reviewed September 5, 2026. Packet: trials/packets/brief-03-attempt-01. All writes are new reviewer artifacts under this report's directory. Packet-relative paths below identify inputs; unprefixed evidence filenames identify files beside this report. No builder was contacted, no other trial or outside builder implementation was consulted, and no consumer render, implementation, recipe or input was changed. This was a review, not a new consumer/build attempt.

Findings and inherited limitations

No new blocking finding

The candidates most likely to invalidate this reuse were checked rather than excused by render success:

• Longer paragraph loses the final lower-line margin: not reproduced. At native L16 / output O41, the custom lower-line ink ends at exclusive x1398, leaving 522 px of right margin; source ends at x1455, leaving 465 px. The custom upper line ends at x1674, so the lower line remains shorter. Both lines survive the longest held state through O82. Bottom/left clipping is the selected treatment and is expressly allowed by the brief; a readable whole paragraph is not required.
• New lead collides with the keyword or the fixed underline misses its emphasis: not reproduced. Independent bundled-font measurement gives Shape our an advance of 569.9285 px in its fixed 586 px box, ending at x973.9285 before the documented x984 limit and the keyword anchor x990. tomorrow measures 540.4992 px, fitting the 555 px box and the recipe's suggested 510–545 px range. It remains a legible, separate emphasized word. The fixed returning gesture is intentionally shorter than the complete word, not an auto-fitted full-width rule.
• Mixed cadence stretches the native treatment or inserts holds: not reproduced. Exact rational conversion, encoded PTS, native render comparisons, and half-native-frame construction checks agree. The only interval extension is the documented 167/24000 s = 6.958333 ms output-grid boundary rounding, not a changed native track or added hold.
• The boundary is a dissolve, matched anchor or blank flash: not reproduced. O82 contains the final white paragraph crop; O83 contains only the incoming cyan layered field. O80–87 show a clean field reset and then Shape at O87. No paragraph residue, black frame or cross-treatment object transition is visible.
• The second shot becomes a static color card or two unrelated straight rules: not reproduced. Corner material, marks and diagonal texture continue moving. The first sweep, opposing return and overlapping retrace remain distinct in source, native custom and output evidence, including the new half-frame samples at 60 fps.

Evidence: paragraph-native-source-edit.jpg, cut-80-through-87.jpg, construction-01.jpg through construction-05.jpg, underline-half-frame-strokes.jpg, underline-hold-exit.jpg, exit-198-through-202.jpg, measurements.json, independent-font-metrics.json, clocks.json, and the moving evidence described below. review.json contains a claim registry with the checked counterevidence.

Inherited: paragraph first-glyph margin and substitute font — remains open, minor

The delivered current report evidence/review-type-a-trio/retest-review01/type-paragraph-cuts.md leaves PC-03/P3 open; it does not grant acceptance. Its default has first-state left ink at x581/581 versus source x562/567. The exact-copy custom result has x571/581, or +9/+14 px, at L0. The different initial C changes the sidebearing compared with the default's E; it does not prove a new camera displacement. The rounded C also reaches y509 rather than the source's y513 cap top. In the final crop, the new q descender reaches y720 versus the source's y682, while the upper line still starts at y489 and the lower at y931.

These are visible substitute-glyph differences, not recovered source font metrics and not silently accepted as exact fidelity. Complete recipe equality confirms unchanged camera, size, tracking and line-height geometry. The earlier material lower-line-footprint problem is not present in this copy; that is separately demonstrated by the native measurements above. This review does not close PC-03 or approve the font substitution on the user's behalf.

Inherited: underline source-review scope — bounded, not full optical equivalence

The delivered evidence/review-type-b/retest-review02/study-keyword-double-underline/review.md has a bounded no-open-findings disposition for restored filled-corner coverage, word/character cadence, returning/retracing strokes, independent field movement and exit. It explicitly does not establish identical source silhouettes, formulas, physical shading, font metrics or precise path curvature. Recipe/artwork notes still call these draft substitutions. I did not convert that report into a blanket fidelity pass.

Fresh comparisons show the same distinction: the source's scientific markings and detailed translucent material are replaced by the shipped drafting marks and softer original folded SVG surfaces. Their identity, texture and optical shape differ materially from the source. They are inherited documented substitutions, and the brief specifically requests the bundled original artwork rather than source-film pixels. The consumer retained that artwork and its motion; it did not introduce a new flattening or replacement.

The gesture is also an original fitted path, not a recovered source stroke. For example, at L13 the source's thresholded first dash occupies x1016–1133/y608–612, versus custom x1018–1132/y608–610 (exclusive bounds). At L24, source and custom overdraw envelopes differ by a few pixels in height/extent. These differences are disclosed; matching sweep/return/retrace timing is not a claim of identical stroke optics. This trial pass does not erase those source-recipe limits.

Independent binding and brief checks

The mechanical audit, finalized at 2026-09-05 21:04:50 UTC, reports mechanicalPass: true, no immutable changes and zero protected watcher events. Its trialPassed: false explicitly awaits a moving/brief review; it is neither an automatic pass nor a discovered visual failure. The audit itself was pinned and left untouched.

Independent checks establish:

• All 6,674 original packet-baseline files match their declared bytes/hashes. This was recomputed against the review's initial snapshot, not copied from the audit.
• Before/after inventories cover 20,485 regular packet files, 2,418,629,729 bytes, including installed packet dependencies. No bytes, file membership or inventoried symlink targets changed. .cache and .git are excluded and were not review inputs. The audit hash also remains unchanged. See pin-before.json, pin-after.json, pin-comparison.json, and supplemental-checks.json.
• Both complete embedded Recipe objects equal the delivered recipe JSON, recursively, including source facts, all tracks, fields, defaults, observations and limits. This is not just matching IDs, recipe hash strings or selected layer values. Native/custom/edit bound recipes and content also agree. See independent-binding-check.json.
• Active content changes are only paragraph, lead, and keyword. The exact paragraph newline is preserved; Shape our is the two-word lead and tomorrow is the eight-character, one-word emphasis. Default phrases remain only as recipe metadata, not displayed content.
• Both before/after hold counts are explicitly zero. The join is kind: cut, continuity: reset, with an empty equalSlots list and an explicit field-reset reason. This follows the delivered README's EditInput API and src/types.ts:50, not an invented assembly format.
• The four original SVG media slots, three gesture paths, drafting-mark path, colors and bundled Geist bytes remain unchanged. SVG parsing finds no image/text/video/audio/script/foreign-object elements or external asset references. Asset declarations are bound, but their legal/authorship truth is not independently certified.
• Fresh current packet CLI verification passes for all three consumer movies and both delivered default movies used in comparison. Independent checks also rehash actual code files, implementation digest, used media assets, complete bound input props and output/manifest bytes. Verification still reports fidelity as not assessed; the moving review is separate.
• Catalog logs show broad discovery followed by the crop and underline queries. The failed phrase query was followed by a successful narrower query; it was not treated as absence. The chosen recipes fit the editorial requirements and documented direct-content fields. Commands, input generator, result and disclosures were inspected. No implementation-edit or undocumented-help evidence was found; logs cannot prove that no unrecorded outside communication ever happened.

Actual consumer outputs

The main MP4 SHA-256 is e9b1f59cf542e251f8e40f513c68e947023a3109395787f381522488bc1d9591; its manifest SHA-256 is 150af4d1da3fe458a243ae6ef66edcd004e96856f3abb6195e6072a824bc6ee8. All other complete digests, including selected source evidence, inputs, recipe bytes and code, are in evidence-ledger.json and the before/after inventories.

Exact clocks derived in this review

All frame numbers are zero-based. L means recipe-local native frame; O means output frame. Source bounds are inclusive/exclusive.

The native sum is 27011/8000 = 3.376375 s. Cumulative rounding to the 60 fps grid gives round(82.5825) = 83 and round(202.5825) = 203, hence 203/60 = 3.383333… s. No inference from the rounded container-duration string is needed. Every decoded frame PTS in the nine inspected input movies lies on that movie's rational cadence.

The actual recipe sampling is:

Paragraph: min(32, O × 400/1001),                 O = 0…82
Underline: min(59, (O − 83)/2),                  O = 83…202

Paragraph cuts are hold steps, not interpolation. Native holds 7/9/17 frames become output runs 18/23/42 frames, with cuts at O18 and O41. Their ideal native cut times are 7007/24000 and 16016/24000 s; their first displayed output samples are 18/60 and 41/60 s, delayed 8.041667 ms and 16 ms respectively. The source clock is not rounded to 24 fps, and the treatment is not stretched to an arbitrary duration.

The outer cut is O82 → O83, at 83/60 s. The incoming text-free O83–86 is native L0–1, not an added opening lead-in. The entire passage begins immediately with paragraph type at O0.

Complete words remain for 46 native frames / 1.533333 s; the completed retrace remains for 36 native frames / 1.2 s, with the field still moving. O201–202 occupy the ordinary final native-frame interval. There is no extra end card, fade, audio or post-treatment hold.

The delivered original-film PTS lists have quantization errors up to 0.708333 ms (paragraph) and 0.666667 ms (underline), recomputed from their absolute frame numbers and rational clocks. Those original-film PTS are supplied metadata; the full source films were not available inside this review scope. The excerpts themselves independently probe as 33/60 frames at the exact nominal rates. Their full-resolution source PNG mappings and hashes match the delivered comparison records. After downsampling those native PNGs, excerpt-frame RGB differences are small (maximum mean 0.704/255), consistent with the stated re-encoding, not frame-index substitution.

Moving and native-frame review

T3 preview_status and then preview_open both explicitly returned no preview automation host available. I used the permitted system Chromium with the existing local Playwright package. The server exposed only allowlisted packet movies/font and reviewer-generated comparisons. Browser profiles, caches, temporary data, screenshots and logs all stayed in this evidence subtree. A short /proc/<reviewer-pid>/fd/<directory-fd> alias to the same allowed temporary directory avoided Chromium's Unix-socket pathname limit; no external temporary directory or builder code was needed.

Moving sources were actually played, not inferred from FFprobe or contact sheets:

• The actual 60 fps consumer edit, both native custom movies, both delivered moving source excerpts and both delivered source/default comparison movies were played from start to end at 1×.
• New synchronized native-clock triptychs place delivered source, delivered default and actual native custom side by side, played at 1× and 0.25×.
• assembly-review-comparison.mp4 places the source passage, every actual consumer output frame, and native custom passage side by side on the independently derived 203-frame clock. It was played at 1× and 0.25×. Source/native panels hold the preceding native sample at fractional positions; the middle retains the actual 60 fps half-native-frame motion. These are labeled reviewer-only comparison conversions, not altered deliverables or claimed original 60 fps sources. Exact mapping is in assembly-comparison-mapping.json.
• Advancing playback screenshots were visually inspected alongside indexed native construction, cut, hold and exit evidence. The source and shipped default are distinguishable from the consumer copy in the screenshots; source text/branding is confined to reviewer evidence and does not appear in the consumer movie.

Final playback logs: playback-final.json and playback-assembly.json. All 14 final passes reached the end. The actual edit's 1× pass has 140 callbacks with zero reported dropped frames; callbacks are not a complete frame census. Each slow native triptych reports one dropped frame, and the 1× assembly comparison reports one. The actual edit and assembly comparison at 0.25× have 202 callbacks and zero reported drops. These limitations are retained, not converted into a zero-drop/continuous-viewing claim.

Independent decoding retains 389 full-resolution frames: all 203 output frames, all 93 native custom frames and all 93 delivered default frames. All 93 native custom positions were numerically compared with the derived matching output positions; the maximum whole-frame RGB mean absolute difference is 0.6374/255. This supports render/clock agreement despite separate lossy encodes; it is not a perceptual fidelity score.

Visual native checks cover paragraph L0/6/7/15/16/32; underline L0–24 and L30/46/57–59; actual cut O80–87; every half-frame stroke state O108–132; and exit O198–202. Not every full-resolution output frame was individually viewed. The final output frame and sampled moving comparisons preserve the complete line and emphasis. Native corner/field measurements between O131 and O201 show continued change in all four independent regions, consistent with visible travel rather than a frozen card. The custom/default artwork regions remain closely aligned at matching native frames.

The paragraph remains emphatic rather than paragraph-wide readable, as requested. The second shot gives a readable complete phrase, separate lead/emphasis, a visible return/retrace, and an uninterrupted finish. No new material mismatch was found in these checks.

Method limits and gate separation

• No continuous-human-viewing claim. This was machine playback plus inspection of advancing samples, actual decoded frames, source-native evidence, measurements and code/input contracts. Browser screenshots are not exact-frame timestamps; indexed decodes resolve the critical moments.
• Delivered source excerpts are 960×540 H.264 re-encodes; retained source PNGs are 1920×1080. The full source films and original font/material construction were not fetched or consulted. Source-file origin and original-film PTS cannot be independently authenticated from this packet alone.
• Contact sheets and reviewer comparisons are scaled/re-encoded. Measurements use full-resolution decodes with stated thresholds; ink bounds are not recovered CSS, vector bounds or proof of exact font weight.
• The first browser launch failed on a long socket path and is retained in playback-run.stderr.txt. An initial successful pass serialized prototype-based quality counters as empty objects; its counters are not used. The final passes explicitly serialize those fields. These are reviewer-instrumentation limits, not consumer defects.
• Manifests, packet baseline, watcher report and logs are unsigned local evidence. The before/after check cannot establish an adversary-proof history or rule out undisclosed communication. Installed dependency bytes are pinned for this review interval, not authenticated as the original render process's dependencies.
• Rights declarations, broad library completion, additional content variants, final playback-platform behavior and user creative acceptance are outside this verdict.

Disposition: trialPass: true; zero new material or minor adaptation findings; inherited PC-03 remains open; source optical/substitution limits remain disclosed; user acceptance is not approved; completed-example credit is 0; publication is not approved. No packet audit, catalog, implementation, input, output or prior review was updated.

Evidence navigation

• review.json: machine-readable verdict, gates, claim registry and coverage.
• evidence-ledger.json / evidence-ledger.md: input/evidence groups, exact before/after hashes and claim links.
• pin-before.json / pin-after.json / pin-comparison.json: full packet byte inventory and unchanged result.
• independent-binding-check.json, supplemental-checks.json, verify-*.json: independent equality, baseline/assets checks and current CLI results.
• clocks.json, probes.json, native-output-agreement.json, measurements.json: exact clocks, encoded PTS, paired frames and native fit/motion measures.
• playback-final.json, playback-assembly.json, playback-final/, playback-assembly/: final moving playback records and samples.
• paragraph-review-triptych.mp4, underline-review-triptych.mp4, assembly-review-comparison.mp4: reviewer-only moving comparisons.
• artifact-inventory.json: hashes of reviewer evidence and scripts, excluding operational browser caches/runtime and the inventory itself.