No blocking proof findings remain for the requested R2/R3/R5 recheck of [#24658](https://git.cafe/pingdotgg/t3code/pull/24658). The previous UUID/address-noise objections are resolved. Reviewed baseline [36d4d711b79c](https://github.com/pingdotgg/t3code/commit/36d4d711b79c73eb33131944f73ea4a8c6156484) and candidate [adf56b5b8a2e](https://github.com/pingdotgg/t3code/commit/adf56b5b8a2ee2f003aa53e705604cc0e26dead3).

The new delivered PNGs independently recompute to the claimed counts. Each of the ten delivered Diff0/DiffT images exactly matches the independently generated decoded RGB image. Every sheet contains the corresponding unaltered before/after/diff panels. R2, R3 and R5-B have no changed pixels anywhere. R5-A/C have zero changed pixels outside x=901–1439, y=125–999 at either threshold; both changed bounding boxes equal that preview viewport. No masking, cropping, alignment correction or excluded region was used in the recomputation.

| Row | Status | Assessment |
|---|---|---|
| R1 | SUFFICIENT, carried forward | Previous accepted four-condition proof and narrow exact reproduction retained. Headline block, recording block, visible-observation line and their links are unchanged. The earlier wide replay reproducibility caveat remains carried forward. |
| R2 | SUFFICIENT | Diff0 / DiffT: **0 / 0**. Recordings show + → Files → scheme.html selection → the file-preview browser-opening action, followed by the real workspace asset page. Candidate outgoing request and loaded snapshot contain explicit Light; baseline omits it. Settled fixture and address bar are identical. |
| R3 | SUFFICIENT within the stated native-Light scope | Diff0 / DiffT: **0 / 0**. Recordings show creation of a real terminal, printing the fixture URL, and clicking that recognized terminal link. Candidate request/snapshot contain explicit System; baseline omits it. Prior mocked initial-reset mechanism support is reused, not rerun. No native-Dark host, live OS switching or first-flash proof is claimed. |
| R4 | SUFFICIENT under the prior accepted exception, carried forward | The complete R4 block, visible-observation line and opener/child ID paragraph are unchanged. Inherited Dark snapshot proof remains accepted; visible child appearance and window.opener remain UNVERIFIED behind the disclosed Idle overlay. |
| R5 | SUFFICIENT, A/B/C | A Diff0 / DiffT: **471,622 / 470,934**; B: **0 / 0**; C: **471,622 / 470,941**. All A/C changes remain inside the preview viewport. Recordings show disabled Settings readbacks Dark → Light → Dark, three real + → Browser creations, and revisits with original candidate appearances Dark / Light / Dark. |

The fixed tokens preserve the requested R5 claim. A/B/C have distinct displayed case labels and three distinct tab IDs, retained from birth through revisits. Snapshots retain A Dark after saving Light, and B Light after saving Dark; final snapshots retain all three tabs and their original preferences. The continuous recordings show the corresponding appearances and zero media-change counters. The fixture still uses matchMedia and CSS media queries; its fixed labels do not encode appearance. I treated those labels as case labels, not independent proof of a unique document-load UUID. Neither that distinction nor the nondisplayed UUID weakens the requested distinct-tab appearance-retention proof.

Opened and visually inspected all **10 new published stills, 10 diff images, 5 comparison sheets, and 6 uploaded contact sheets**, plus the four additional final-run A-first/B-final local stills. Independently decoded all **six MP4s** with ffmpeg and inspected **122 frames at approximately 1 fps in 16 contact sheets** across their complete durations. Additionally inspected the fast R2 Files opening at 25 fps and all six R5 Browser creations at 10 fps. The timestamped narration/action tables agree with the recordings, allowing ordinary frame quantization. These MP4s have no audio narration track.

Settled proof has no ghosting, stale overlays, loading indicators, clipped fixture text, wrong theme or wrong viewport. Recordings honestly include temporary Connecting/loading states, Settings transitions and deliberately opened Appearance menus, which clear before settled captures. Ordinary tab-title/address-field truncation does not obscure the proof. R5 fixture-API preparation captions are labelled evidence annotations. The opening paths are UI-driven in both the recordings and the matching driver; production settings API preparation is not substituted for preview creation.

R1's headline and recording blocks and the R4 block are byte-for-byte equal to the supplied older Evidence section, with matching SHA-256 hashes saved in `carried-block-comparison.json`. The old matrix is a planning document at an older pair; new artifact labels, launcher checks, recorded checkout receipts and the independently verified current clean checkout heads all identify the requested pair. Both runs' dev-runner logs confirm identical ports 6733/14773. The uploaded scripts match the inspected local scripts; all 31 published PNGs and all six MP4s match their local final-run artifacts.

Audit outputs: [recomputed counts and exact diff/sheet checks](/tmp/lead-24658-r1/proof-audit-3/recomputed-summary.json), [unchanged-block hashes](/tmp/lead-24658-r1/proof-audit-3/carried-block-comparison.json), [retained tab IDs/preferences](/tmp/lead-24658-r1/proof-audit-3/persistence-check.json), [recording decode inventory](/tmp/lead-24658-r1/proof-audit-3/decoded-recordings.json), [checkout verification](/tmp/lead-24658-r1/proof-audit-3/audit-final-verification.json).

Read-only audit: no application edits, pushes, comments, servers, browser sessions or test runs. All audit writes are under the requested scratch directory; nothing was deleted. This is an evidence recheck, not a new full code or merge-readiness review.

FRICTION:

infra | Seven initial IPv4 artifact downloads failed DNS resolution; curl retries succeeded | Resolve the artifact host reliably, or use bounded retries with --retry-all-errors.

VERDICT: SUFFICIENT