# Evidence packaging correction — revision6 Issue [#14743](https://github.com/pingdotgg/t3code/issues/14743), PR [#22208](https://github.com/pingdotgg/t3code/pull/22208). Metadata-only addendum to the existing partial delivery. No source, fixture, production, worktree, browser, server, original screenshot, original recording, or original capture metadata was changed. The independent remedial capture agent's work is outside this package and has not been assessed here. Read `pixel-manifest-corrected.csv` for all300 checkpoint classifications and concrete nonzero observations; `action-index.md` / `.json` for36 content-based action windows; `counts.json` for machine-readable totals; `source-sha256.json` for original asset hashes. Supplemental inspection sheets are in the archive. They crop/scale originals solely to inspect visible differences: **the original full-surface pixel counts were preserved without masking, registration, crops, or alignment.** ## Accurate counts, narrower meaning 128 physical pairs /172 unmatched, by combination:1440 light41/34;1440 dark,390 light,390 dark29/46 each.12 zero pairs and116 nonzero. Original independently audited totals retained: diff0 **13,330,450**, diff16 **9,810,546** physical saved-image pixels. These measure saved-image delta; they do not validate equivalent interactions. `asset_status` preserves whether both files exist; `comparison_class` separately states whether the comparison is usable. `valid_paired` means retained same-named paired output/divergence evidence, not identical input focus, real persistence, complete wire logs, or an assertion of zero incidental differences. `baseline_unreachable` means the exact authoritative-null draft interaction cannot exist on the immediate-write baseline; similar final persisted outputs can still be reachable. `not_captured` never means impossible. `unequal_seed` includes unestablished/retained baseline values; per-row reasons identify actual text, hydration and theme mismatches. No synthetic counterpart was created. | Asset availability | valid_paired | baseline_unreachable | not_captured | unequal_seed | |---|---:|---:|---:|---:| | paired_assets | 90 | 2 | 5 | 31 | | unmatched_asset | 0 | 58 | 114 | 0 | Five physical baseline R2 pairs are now `not_captured` with `INVALID_BASELINE_ENTER`: Enter hit the baseline switch. Three R3 pairs and all28 project pairs are `unequal_seed`; two R4 draft pairs are `baseline_unreachable`. The counts are retained even for invalid comparisons. Equivalent persisted8 output can be viewed in original `I-on8` and harness `H-baseline-ack8` with their original provenance, but neither substitutes for the missing baseline input-Enter interaction. ## What inspection actually found All116 nonzero pairs were inspected in supplemental before-left / after-right sheets. Every row now describes observed differences, rather than “may differ.” Examples: baseline830 versus after30 in edited/saved fields; baseline80 versus after0; baseline83650 versus after3650; focused after input8 versus neutral baseline input8; baseline Off versus after On8 in transitional browser/project frames; Delete now enabled/Last run just now/1 kept versus disabled/Turn on a cleanup rule/0 kept; switch track/thumb differences during Off transitions; harness before/after heading, source SHA and request1=8 versus empty request log; sidebar, row displacement, and loading/T3-mark differences. Desktop dark project baseline images are light while after images are dark. Mobile project images in the group labeled dark are actually light on both sides. These assets remain available, but do not establish a correct project dark-theme matrix. Baseline project return stills show the loading surface on desktop; they are not loaded destination comparisons. The1440dark I-off-again frame also includes after settled8 versus baseline settled Off. Labels alone cannot certify settled states. This is a bounded visible-region accounting, not an explanation of every physical changed pixel. Desktop dark layout displacement (sidebar boundary, row/text positions), narrow vertical stripe pixels, subthreshold deltas and remaining fine glyph/transition edges retain unresolved attribution. None is waived as “just AA,” and this package does not claim exhaustive pixel-identical proof. The original full images/diffs remain the inspection authority. ## Replay correction The auditor's one pending replay is semantically reproduced, not pixel-identical: diff0=11,698 /diff16=7,797. Heading, paragraphs, callback label, row label and button text edges differ; log content, borders, switch and layout match. The **28 threshold pixels at x1126..1127/y198..211 are a real caret present only in replay: capture timing, not antialiasing**. No new replay was run here. ## Remaining proof gaps This metadata does not repair the independently identified R1 bounded request windows/Delete-now framing; R4 actual Tab/real days and row-gap clicks; R5 empty outside blur; R7/R10 actual scoped runner failure/toast/unavailable/partial result; R8 real second submission before old failure; R11 correct outside release target/event logs. Old R8 success was injected by unknown-ID resolve and remains invalid. Old R11 release targeted inner Increment and remains invalid. R9 controlled ordered callback evidence retains its original limited meaning; no transport reorder guarantee is added. The36-window action index gives35 content-indexed entries and1 explicitly labeled video gap:1440light project return-to-Storage still has no verified matching frame in the original18.481s raw segment. The video shows departure to General; destination readback remains separate still-only evidence. Missing startedAt values remain null. No action timestamps or missing proof were fabricated. ## Existing public assets — preserved [after-1440-dark.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/eda58663ed09e73a/after-1440-dark.mp4) [after-1440-light.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f33c7664e71f1a9e/after-1440-light.mp4) [after-390-dark.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7f97d88c456ccd5a/after-390-dark.mp4) [after-390-light.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/dd07b6282c567a70/after-390-light.mp4) [before-1440-dark.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a44a8d61f855aad7/before-1440-dark.mp4) [before-1440-light.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/82fc0b7e1b7d290e/before-1440-light.mp4) [before-390-dark.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/139b887c22ddb4b8/before-390-dark.mp4) [before-390-light.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d6648820fdaa3cd1/before-390-light.mp4) [stills-1440-dark.zip](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6676132feeb595c2/stills-1440-dark.zip) [stills-1440-light.zip](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8c5b57d32044a9d5/stills-1440-light.zip) [stills-390-dark.zip](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/af6dc00324785ab9/stills-390-dark.zip) [stills-390-light.zip](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/1bf4ac2ece8d04d3/stills-390-light.zip) [raw-recordings.zip](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/da388049d258c808/raw-recordings.zip) [manifest-and-tooling.zip](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8341c27a07bab8f6/manifest-and-tooling.zip) ![overview.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4394d9d2d1b79187/overview.png) [index.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/878f173ce72ae3de/index.md) ## FRICTION - No browser/server was opened. Only read-only offline source reads, ffmpeg/Python inspection, and writes under `/tmp/issue-14743-metadata-r6` were used. No forge writes, push, watch, or production/fixture edits. - A scratch file initially named inspect.py shadowed Python's standard-library inspect module; renamed locally. A cleanup command was rejected by the command runner; no source data was removed or modified. - Full-surface160x100 nearest-frame matching confused repeated8/Off states. Its candidate JSON is retained as exploratory data, explicitly not authoritative; the published action index uses inspected chronological raw-video content instead. - Raw videos have no per-key overlay, so visible content cannot independently prove Enter/Space/Tab. Script provenance is separate from observed output. One return-to-Storage video mapping remains unverified rather than invented. - Transition frames, baseline appended8 text, theme mislabels/hydration and layout displacement limit comparison validity. Fine pixel attribution remains open; no exhaustive-proof claim. - Uploads use the [ghdrop skill](/home/ubuntu/.codex/skills/ghdrop/SKILL.md). No PR or issue comment is posted by this task. ## Corrected public downloads - [metadata-r6.zip](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2e17bed53a1a0c75/metadata-r6.zip) - [pixel-manifest-corrected.csv](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2b00b21e598a7f2f/pixel-manifest-corrected.csv) - [action-index.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5d00d240b878c38d/action-index.md) - [action-index.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5b538e0102ceeef5/action-index.json)