# W10 Evidence — round3
**W10 — real paired environment and repository switches, round 3**
Before [15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25); after [597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e), the exact requested revisions for [#21631](https://git.cafe/pingdotgg/t3code/pull/21631). **New capture scope: W10 only.** This replaces round2’s rejected hook-stub/handmade-identity evidence. Prior tab, cached-detail and thread-surface subcases remain below; this run does not re-audit those other recordings.
**Production boundary:** two independent real servers per revision, each with a fresh home and `T3CODE_DEV_SEED=0`. The web client is paired to both real environment identities. E1 uses the normal app pairing route; E2 is registered by the app’s actual `connectPairing.run(appAtomRegistry, {pairingUrl})` command, the same command AddEnvironmentDialog calls. E2 pairing is invoked programmatically; no UI pairing-dialog click claim is made. Both prepared connections and their real destinations are recorded. This is not the transport-only fallback: no asset hook, atom family, RPC client, connection runtime, WebSocket payload or endpoint is replaced.
The labelled PR-view input harness mounts production PullRequestMarkdown, its real repository context, production image/video producers and their actual web asset hooks. It keeps the PR media URLs and `/synthetic/personal-fork` cwd fixed. The reused round1 proxy supplies synthetic successful media GET bytes and a synthetic PR-detail endpoint; this W10 harness supplies its PR metadata directly to the production markdown context rather than mounting the whole panel. Discovery, pairing/session HTTP, asset-ticket HTTP and WebSocket RPC remain production. A passive WebSocket send observer calls the native send unchanged and records exact request resources, successful responses, decoded synthetic signed claims and capability hashes. Signed capability URLs and credentials are redacted. This proves asset routing/cache identity, not upstream GitHub fetching/authentication or full-screen navigation.
**Actual cache identity:** the observer correlates each displayed image/video capability with its real RPC request, obtains `assetEnvironment.createUrl({environmentId,input:{resource}})`, assigns a WeakMap ID to that actual atom object, reads its production `.label`, and verifies its cached successful response hashes to the displayed source. IDs below compare object references; they are not a driver-calculated canonical resource key. Exact production labels and real environment UUIDs are retained in each transcript.
| Side | Alias | Real environment ID | Browser RPC endpoint | Real server port |
|---|---|---|---|---|
| before | E1 | `75db8def-d97f-4ab4-8c94-eed1a9775441` | `ws://localhost:19431/ws` | 15558 |
| before | E2 | `c151125a-c260-423a-918b-835406cf5347` | `ws://localhost:19432/ws` | 19332 |
| after | E1 | `7aad7973-45b8-4837-935f-37f483efa809` | `ws://localhost:19441/ws` | 16258 |
| after | E2 | `1742c5ae-7dad-40e3-99c4-fb5ac4261411` | `ws://localhost:19442/ws` | 19342 |
Proxy WebSockets forward to the actual server (E1 through Vite’s `/ws` proxy); E2 uses its own socket and independent server/signing home. These are two local environments, not two machines or a tunnel/relay test.
**Result:** base omits `repository` on every real request and signed claim. Head carries B (`fixture-org/media-upstream`) at E1 and E2, then C (`fixture-org/media-alternate`) after changing only the PR repository. The same image and video URLs persist throughout.
| Settled step | Before: image/video atom objects; requests total | After: image/video atom objects; requests total |
|---|---|---|
| E1 / B | atom-2/atom-1; 2 | atom-2/atom-1; 2 |
| E2 / B | atom-4/atom-3; 4 | atom-4/atom-3; 4 |
| return E1 / B | atom-2/atom-1; 4 | atom-2/atom-1; 4 |
| refresh E1 / B | atom-2/atom-1; 6 | atom-2/atom-1; 6 |
| E1 / C | atom-2/atom-1; 6 | atom-6/atom-5; 8 |
| return E1 / B | atom-2/atom-1; 6 | atom-2/atom-1; 8 |
E2 produces distinct objects and two RPCs in both revisions. Returning to E1 restores the original successful objects without a redundant mint. The explicit Refresh selected assets action calls the real registry refresh on those production atoms, producing two more RPCs to E1. B→C leaves base’s objects unchanged because repository is absent from its resource. Head creates two new C atoms and requests; C→B restores the original B objects without a new RPC. There are six successful base requests/replies and eight successful head requests/replies, with no failures in the accepted sequence.
**Same driver and narrated sequence:** E1 B → E2 B → E1 B → refresh E1 → repository C → repository B. The movies add spoken step narration and an elapsed timestamp footer to the original Preview recording. Original raw recordings are also uploaded. Narration is an editorial layer over the captured actions; every raw frame and its presentation timestamp is retained (114 before,126 after), with an elapsed footer in the blank lower area. Variable-frame-rate track duration metadata differs after encoding; the raw originals preserve the source bytes. The table gives driver action→settled-observation timestamps relative to its recording-start request (subsecond recorder/tool latency is not a performance measurement).
| Step | Before action → observation | After action → observation |
|---|---|---|
| E1 B | 1.048s → 1.118s | 0.972s → 1.120s |
| E2 B | 4.830s → 5.326s | 5.057s → 5.595s |
| return E1 B | 9.339s → 9.809s | 9.572s → 9.978s |
| refresh E1 B | 13.836s → 14.459s | 13.885s → 14.336s |
| repository C | 18.960s → 19.367s | 18.179s → 18.704s |
| return repository B | 23.155s → 23.649s | 22.585s → 22.992s |
| Before recording | After recording |
|---|---|
| [w10-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b8876a797f4ddd98/w10-before.mp4) | [w10-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/916d0bb032e1d574/w10-after.mp4) |
Raw originals: [w10-before-raw.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d24cc30dfe9f4ff2/w10-before-raw.mp4), [w10-after-raw.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f9f83d646b426661/w10-after-raw.mp4). Raw RPC transcripts: [w10-before-rpc.jsonl](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/130c50bc1723b2d5/w10-before-rpc.jsonl), [w10-after-rpc.jsonl](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/33128d2437ec20fa/w10-after-rpc.jsonl). Complete state/cache observations: [w10-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/04142cc3e85d5db2/w10-before-transcript.json), [w10-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e1beaf1ba66f7294/w10-after-transcript.json).
**Rendering and settled frames:** Linux x86_64, T3 Preview Chromium156.0.8078.70, no device emulation, DPR2, font scaling100%, root16px, dark appearance, CSS viewport1440×1000. Preview saves full-viewport PNGs normalized to1280×889; the original/final recordings are2880×2000. Video is paused at0.1s; fonts, image decode and media seek are awaited. Each accepted still has a second capture500ms later with pixel-identical results. All24 captures and12 diffs were opened and inspected, as were the six review montages and chronological decoded recording frames. No clipped text, duplicate overlay or loading indicator appears in the settled proof. During recorded transitions, production loading can briefly appear and the diagnostic readout is refreshed at the settled observation; those earlier frames are not resource proof. Playback/control timing can differ between the two recordings.
Diff0 multiplies each nonzero channel delta by255 and saturates. DiffT highlights a per-channel delta strictly greater than8/255. Differences are confined to the harness repository readouts; the C state also changes atom IDs and request count, and the final B state retains the cumulative count difference. The production media/layout area is pixel-identical in each comparison. These are intentional diagnostic differences, not a product layout change.
**e1-initial:** 2,887 changed pixels at threshold0; 2,422 at >8/255, out of1,137,920.
| Before [15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25) | After [597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e) | Diff0 ×255 | DiffT >8/255 |
|---|---|---|---|
|  |  |  |  |
**e2:** 2,887 changed pixels at threshold0; 2,422 at >8/255, out of1,137,920.
| Before [15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25) | After [597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e) | Diff0 ×255 | DiffT >8/255 |
|---|---|---|---|
|  |  |  |  |
**e1-return:** 2,887 changed pixels at threshold0; 2,422 at >8/255, out of1,137,920.
| Before [15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25) | After [597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e) | Diff0 ×255 | DiffT >8/255 |
|---|---|---|---|
|  |  |  |  |
**e1-refresh:** 2,887 changed pixels at threshold0; 2,422 at >8/255, out of1,137,920.
| Before [15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25) | After [597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e) | Diff0 ×255 | DiffT >8/255 |
|---|---|---|---|
|  |  |  |  |
**repo-c:** 3,176 changed pixels at threshold0; 2,597 at >8/255, out of1,137,920.
| Before [15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25) | After [597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e) | Diff0 ×255 | DiffT >8/255 |
|---|---|---|---|
|  |  |  |  |
**repo-b-return:** 2,958 changed pixels at threshold0; 2,467 at >8/255, out of1,137,920.
| Before [15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25) | After [597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e) | Diff0 ×255 | DiffT >8/255 |
|---|---|---|---|
|  |  |  |  |
Validation and reproduction: [w10-metrics.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/95b83a257f780574/w10-metrics.json), [validation-results.txt](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7060bccffd1271ed/validation-results.txt), [REBUILD-round3.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/38b0809bf0c3ed7f/REBUILD-round3.md), [fixture-proxy-round3.mjs](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/99956b84a5f55733/fixture-proxy-round3.mjs), [w10-runtime-harness.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f8013a4dbc2a1611/w10-runtime-harness.js), [capture-w10.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5ef2286ccce17442/capture-w10.js), [analyze-w10.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5392135238c17fa3/analyze-w10.py), [narrate-and-review.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0e8193a112e252bc/narrate-and-review.py), [w10-recording-validation.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d1b0c7c15ac83efc/w10-recording-validation.json). All repeat captures and scripts are included in the verbatim upload index below.
**Retained round1 subcases and tables**
Round1 main recording14s B→C,15s C→B with fixed URL/cwd,25–26s Summary→Timeline→Summary. Panel-state footage covers cached refresh; thread-surface footage0s unknown capability,2s unsupported,4s supported,6s close,8s reopen.
| Before
[15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25) | After
[597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e) | Diff0
[15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25) → [597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e) | DiffT >8/255
[15707a84](https://git.cafe/pingdotgg/t3code/commit/15707a84b03e27bb257d60e027812ebf94bbda25) → [597f1aea](https://git.cafe/pingdotgg/t3code/commit/597f1aea6191d434ad23877e0e94121b7858382e) |
|---|---|---|---|
| [interactions-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/26fea6363c29fe51/interactions-before.mp4) | [interactions-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/1de08805e0a37ec0/interactions-after.mp4) | N/A: recording | N/A: recording |
| [panel-states-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/1d4652335483d24b/panel-states-before.mp4) | [panel-states-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e2fa3f5bb8c159d7/panel-states-after.mp4) | N/A: recording | N/A: recording |
| [thread-surface-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2446664f590de68f/thread-surface-before.mp4) | [thread-surface-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fa4b53d362eba472/thread-surface-after.mp4) | N/A: recording | N/A: recording |
**Cleanup**
All task-owned dev/API/proxy listener ports are closed, the owned Preview tab is closed, the before checkout is removed, and the after checkout remains clean at the requested head. Scratch is retained. Final free disk:17GB. No PR body, source file, commit or live database was changed. [cleanup-complete.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3bee9afb74a66aa3/cleanup-complete.json).
**Upload output, verbatim**
The current links above use the final upload. Earlier duplicate filenames in this chronological upload history are superseded by the last occurrence.
[REBUILD-round3.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/38b0809bf0c3ed7f/REBUILD-round3.md)
[analyze-w10.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5392135238c17fa3/analyze-w10.py)
[build-reports.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/566d7f1cf2912e2a/build-reports.py)
[capture-w10.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5ef2286ccce17442/capture-w10.js)
[cleanup-complete.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3bee9afb74a66aa3/cleanup-complete.json)
[cleanup-owned.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b1ddcec3de11cf2e/cleanup-owned.py)
[cleanup-signals.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/9a939240b37adcbc/cleanup-signals.json)
[fixture-proxy-round3.mjs](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/99956b84a5f55733/fixture-proxy-round3.mjs)
[narrate-and-review.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/feed6b975c9921bc/narrate-and-review.py)
[upload-artifacts.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bed6cebc6a9feb2b/upload-artifacts.py)
[validation-results.txt](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7060bccffd1271ed/validation-results.txt)
[w10-after-narration.wav](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5484a3fc3dbf5241/w10-after-narration.wav)
[w10-after-raw.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f9f83d646b426661/w10-after-raw.mp4)
[w10-after-rpc.jsonl](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/33128d2437ec20fa/w10-after-rpc.jsonl)
[w10-after-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e1beaf1ba66f7294/w10-after-transcript.json)
[w10-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f6f42f8938428960/w10-after.mp4)
[w10-before-narration.wav](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a16c4533ddb99a78/w10-before-narration.wav)
[w10-before-raw.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d24cc30dfe9f4ff2/w10-before-raw.mp4)
[w10-before-rpc.jsonl](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/130c50bc1723b2d5/w10-before-rpc.jsonl)
[w10-before-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/04142cc3e85d5db2/w10-before-transcript.json)
[w10-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6d6691cdb9f898c1/w10-before.mp4)
[w10-metrics.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/95b83a257f780574/w10-metrics.json)
[w10-runtime-harness.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f8013a4dbc2a1611/w10-runtime-harness.js)
[narrate-and-review.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0e8193a112e252bc/narrate-and-review.py)
[build-reports.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7b25934436e64648/build-reports.py)
[w10-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b8876a797f4ddd98/w10-before.mp4)
[w10-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/916d0bb032e1d574/w10-after.mp4)
[w10-recording-validation.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d1b0c7c15ac83efc/w10-recording-validation.json)
FRICTION:
- tool | Reused byte proxy intercepted a bearer asset-ticket route; restricted replacement to signed byte GETs and verified real E2 connection/RPC. | The reuse driver should distinguish signed assets from auth-ticket endpoints.
- tool | Preview readiness observer missed a media event during the first attempt; that recording was excluded and both sides rerun with source/decode/seek readiness. | Wait on the rendered source and its own media lifecycle.
- runbook | E2’s startup code expired after5m; minted a fresh code against only its isolated home. | Pair both environments immediately or mint just before use.
- tool | Direct multi-source flite mux stalled; generated finite narration WAVs first, then muxed with an explicit source duration. | Produce bounded audio before muxing.
W10 round3 evidence capture complete; no new full-matrix audit claimed.