# Supplemental evidence — [#32816](https://git.cafe/pingdotgg/t3code/pull/32816)

Pinned real-app comparison: before [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718), after [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b). This packet only supplements rows marked missing/partial in [proof audit pass 1](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e19de3c5ecd60fc6/proof-audit-pass1.md). Retain all previously accepted evidence. No code, PR body, commits, or pushes were changed.

## Scope and rendering conditions

23/26 assigned in-scope supplemental rows captured; two partial (M01 exact light still pair, M09 baseline/platform set), one unverified (X03 full reconnect/environment-switch). There are nine S rows, seven L rows, eight M rows, and X01-native/X03. Row-level qualification below takes precedence over a broad coverage count. X04 and M10 are not captured: out of scope for this PR.

Real web apps run from their own pinned checkouts and isolated fresh databases. Every dev server started with T3CODE_DEV_SEED=0; no live developer database was read or copied. Synthetic fixtures were inserted through typed real-app commands and the free Replay (dev) provider. Both native fingerprints matched across revisions, so cached dev clients loaded each checkout's Metro JavaScript. Ordinary pending, consumption, cancellation, approval, and run completion are real server behavior. H explicitly labels injected client selectors, command defects/results, or promise holds.

Web W = Chromium 157 on Linux x86_64, shared T3 browser preview, 1440×900 CSS pixels, DPR 2, default font/zoom 100%, no device emulation. N = same browser at 390×844 CSS, DPR 2. Whole surfaces are recorded; contact sheets used only for private inspection. Themes are stated per row. The preview tool exposes color-scheme but no reduced-motion emulation; no fake media-query implementation was used. Accepted light L07 stills waited for fonts/resources, blur and pointer removal, then two full PNGs 500 ms apart with exact ImageMagick AE=0 per side. All published captures and both diffs were opened and visually inspected. Unstable native still attempts were excluded.

Android A = leased Pixel 9, API 36, Hermes, window 411.428571×923.428571 logical, DPR 2.625, fontScale 1; screenshots 1080×2424. Screen recordings may encode a scaled 912×2048 full surface. Android reduced-motion animations were disabled through device settings. iOS I = leased iPhone 17 Pro, iOS 27.2, Hermes, 402×874 logical, DPR 3, fontScale 1, full video 1206×2622. Pixel Tablet = real leased Android API 36, whole 2560×1600 output; DPR/logical viewport could not be recovered before worker disconnect, so neither is inferred. Native menus are the real platform renderers. All mobile evidence uses agent-device and cached mobile:install dev clients over Tailscale to isolated servers; no web emulation is relabelled native.

The browser JavaScript clock was frozen at 2026-11-10T12:10:00Z, except the final S04 pair which restored real Date/Date.now so tooltip delays could finish. Server and OS clocks remained live; native status bars and run duration labels consequently differ. Native still normalization animated the charging icon, so no unmatched/unstable stills are presented as a static pair. Interaction recordings retain their genuine clocks and evolving histories. Driver navigation through the app's real root navigation ref was used where deep links did not route deterministically.

## Coverage ledger

| Rows | Supplemental result | Qualification |
|---|---|---|
| S05, S06 | Captured | Labelled hold; one dispatch despite repeated clicks. Existing accepted busy styling remains primary fit proof. |
| S07, S08, S13 | Captured | Labelled Error, non-Error, interruption; real retry. |
| S09 | Captured H | Source ready; target hydration held/released at quiet-thread atom boundary, then real consumption. |
| S11 | Captured | 390 dark recording continues to unobscured reset. |
| S12 | Captured H | Blank/loading and deleted fallback bodies; no claim of cancelling a deleted server thread. |
| S04 | Captured | Real read-only pairing, connection tooltip, mouse/Enter nonactivation. |
| L01K, L05, L06, L07, L08, L09, L10 | Captured | Real Enter, labelled failure, generic graph-shell omission, long/narrow fit, fresh client, held Disconnect, lineage fallback. |
| M01-light | Partial against requested still pair | Light selected→cancel→unselected recording and baseline captured. Exact stable still pair/diffs not supplied; charging-icon instability. Audit permits retaining valid recordings. |
| M03 | Captured | Native before has no cancel while active; after selected→cancel→absent→finish→unselected. |
| M04-finish | Captured continuation | Adds only reappearance after real approval/completion; retain previously accepted selected/cancel/absent recording. |
| M05, M06 | Captured H | Native held retry/repeat. Failure is a client command defect before RPC, not server rejection. |
| M07 | Captured with disclosed fixture differences | Android after initially-null target→selected→consumed absent; native baseline has no item throughout. |
| M08 | Captured | Android dark real keyboard overlay pair. Supplemental light after also supplied. |
| M09 | Partial | Android tablet Files after flow captured; before worker disconnected/lease expired. iOS split not captured within budget. |
| X01-native | Captured continuation | Native before/after target-next-message sequences and receipts; retain accepted web/late-no-op proof. |
| X03 | Not captured in full | Fresh web/native target subscriptions and cross-client consumption captured separately. Reconnect retention and environment-switch isolation are unverified. |
| X04, M10 | Not captured | Out of scope for this PR. |

## S11 — N×dark: unobscured narrow-sheet reset

Steps: start with source pending in the sheet. At ~0.6 s click its header control. After: cancellation clears selected state; allow the toast to expire (~6 s), and keep the unobscured unpressed header visible through the ~14 s ending. Before: the same activation repeats merge; it has no cancel state. Retain already accepted N×light/dark pending stills.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [S11-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/883dc3bd246f449b/S11-before.mp4) | [S11-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/31458311fc035cfe/S11-after.mp4) |



## L01K — W×dark: actual Enter cancellation

Steps: focus the actual Lineage merge icon; named cancel tooltip is visible during the first ~2 s. Send a real browser Enter key at ~2 s. After: icon clears by ~3 s and fork remains selected; blur around ~4.5 s for clean ending. Before: Enter repeats merge and navigates parent.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [L01K-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/47b73e8519ff45ae/L01K-before.mp4) | [L01K-after-final.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/7d8c9fb947834656/L01K-after-final.mp4) |



## L08 — W×dark: fresh direct fork client

Steps: a fresh incognito client starts in Settings with neither source nor target detail loaded. At ~1.2 s route directly to the fork; never select its parent in this client. After target subscription hydrates, the action is selected. Around ~10 s send the target next message from the separate admin browser; the mounted fork icon resets. Before: ordinary unpressed merge control has no pending/cancel styling. This does not establish reconnect or environment-switch isolation.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [L08-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a76fd3f835ab149b/L08-before.mp4) | [L08-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d21e7a95694808f2/L08-after.mp4) |



## L10 — W×dark: full-route side chat with lineage fallback

Steps: create side chat before its parent has a stable run, then finish both. Verify forkedFrom=null. At ~1 s activate Lineage merge; parent opens. Return to source route at ~3 s. Activate again (~9 s after; ~5 s before). After cancels and stays source; before repeats merge and selects parent.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [L10-before.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b4594a315f194429/L10-before.mp4) | [L10-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f8fb375cd3553ba1/L10-after.mp4) |



## S09 — W×dark: source ready, H target hydration hold/release

Steps: source detail is real and loaded; quiet target projection is held at null. At ~1.5 s release the target pending projection; after becomes pressed without activation (~2–4 s). At ~4 s send target next message; after resets by ~7 s. Both sides use the same hold boundary. This is a labelled target-hydration harness on the production side pane, rather than a claim that the parent route was never selected.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [S09-before-hydration-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d5a89fb6c65ca576/S09-before-hydration-harness.mp4) | [S09-after-hydration-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/dae4e5ee01bdeab1/S09-after-hydration-harness.mp4) |

[Hydration and Disconnect observations](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f5c8f483d9285d6b/web-hydration-detach-transcript.json) confirm sourceLoaded=true, pressed false→true→false.

## S07 — W×dark: labelled command result

Error result at ~0 s; retained selection/busy cleared and Error.message visible by ~1.5 s. The accepted clip is trimmed to the first 26 s before the next fixture starts. Faults are injected before RPC; final retry invokes the real original command.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [S07-S13-S08-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6332c9ebc4f47011/S07-S13-S08-after-harness.mp4) |



## S13 — W×dark: labelled command result

Non-Error failure at ~5.4 s; cancel-specific feature title without a description, selection retained. The accepted clip is trimmed to the first 26 s before the next fixture starts. Faults are injected before RPC; final retry invokes the real original command.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [S07-S13-S08-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6332c9ebc4f47011/S07-S13-S08-after-harness.mp4) |



## S08 — W×dark: labelled command result

Interrupted result at ~10.8 s; busy clears without feature success/error toast; real retry at ~16.2 s clears selection by ~18 s. The accepted clip is trimmed to the first 26 s before the next fixture starts. Faults are injected before RPC; final retry invokes the real original command.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [S07-S13-S08-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6332c9ebc4f47011/S07-S13-S08-after-harness.mp4) |



## S05 — W×dark: labelled held command/repeated clicks

Merge held at 0 s; repeated clicks at ~0.7 s yield one dispatch; release ~2.3 s, pressed ~3.8 s. These are repeat supplements to accepted visual busy captures. Synthetic keydown in the driver does not cause browser default activation: only repeated clicks count here. Actual Enter has its own accepted browser recordings.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [S05-S06-held-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/88f44c684a4c216d/S05-S06-held-after-harness.mp4) |

[Held command transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/db0a885bf0cc959d/held-web-transcript.json).

## S06 — W×dark: labelled held command/repeated clicks

Cancel held ~4.5 s; repeated clicks ~5.2 s yield one dispatch; release ~6.8 s, unpressed ~8.3 s without remerge. These are repeat supplements to accepted visual busy captures. Synthetic keydown in the driver does not cause browser default activation: only repeated clicks count here. Actual Enter has its own accepted browser recordings.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [S05-S06-held-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/88f44c684a4c216d/S05-S06-held-after-harness.mp4) |

[Held command transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/db0a885bf0cc959d/held-web-transcript.json).

## L05 — W×dark: labelled cancellation failure and retry

Steps: inject Error result on first cancellation (~1.2 s); selected state remains. Real retry at ~3.6 s clears it while staying fork. The feature error closes on retry; separate generic runtime feedback may remain visible.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [L05-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d7615a53d1062403/L05-after-harness.mp4) |



## S12 — W×dark: H missing source shell

Steps: omit only the source shell; retain real target pending. At 0 s shell snapshot readiness=false shows blank fallback body with Side chat title. At ~1.8 s readiness=true shows deleted body. After remains pressed across both; before remains disabled/unpressed. No cancel is attempted against the deleted-source fallback.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [S12-before-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5df18c4f1ea8f629/S12-before-harness.mp4) | [S12-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2618ab3db45a26a3/S12-after-harness.mp4) |



## L06 — W×dark: H omitted parent graph shell

Steps: omit parent only from relationship graph shells; keep the real target projection and retained title-lookup row. Focus generic action at 0 s; generic source cancel tooltip appears in the opening ~2 s. Click ~2.2 s; reset, stay fork, and blur before ending. Parent status becomes Unknown because its graph shell is deliberately missing.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [L06-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d186a9899b2dc288/L06-after-harness.mp4) |



## L07 — N×dark: long Lineage title, tooltip and navigation

Steps: full-density details popover on long-fork. Focus merge icon at 0 s for wrapped named tooltip. Click ~1.8 s; after cancels and stays fork (~3–6 s). Click distinct parent link ~7.1 s; parent opens by ~8 s. Before activation repeats merge and reaches parent around ~2 s.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [L07-before-390-dark.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/85c92f5f03221211/L07-before-390-dark.mp4) | [L07-after-390-dark.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/07b807e3b0fa23c5/L07-after-390-dark.mp4) |



### L07 — N×light pending stills, whole surface

Before each accepted still: fonts/resources settled, no focused/hovered control, two whole PNGs 500 ms apart with AE=0. Both are 780×1688 physical (390×844 CSS, DPR 2). Diff0 = absolute RGB difference amplified ×32; DiffT = red where any channel differs by >8/255, white elsewhere. All four images were opened before upload.

| Before | After | Diff0: threshold 0, ×32 | DiffT: >8/255 |
|---|---|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) | [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) → [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) | [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) → [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| ![L07-before-390-light-pending.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/70edf33a2439de46/L07-before-390-light-pending.png) | ![L07-after-390-light-pending.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/006670d8f02ecb3b/L07-after-390-light-pending.png) | ![L07-light-diff0-amplify32.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/48f39ffece6be206/L07-light-diff0-amplify32.png) | ![L07-light-diffT-over8.png](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f272d3b404859b06/L07-light-diffT-over8.png) |

What else changed in this static pair: only the intended selected control background and glyph color. Parent text, status, spacing, rest of the whole surface and frozen browser clock are identical.

## L09 — W×dark: H Disconnect held

Steps: actual More thread actions menu opens at 0 s; choose Disconnect ~0.9 s. Hold stopSession command result through ~2.2 s: selected merge control is disabled and retains merge glyph, without merge spinner. Release ~3.1 s to original real stopSession; source remains selected. DOM observation records calls=1, pressed=true, disabled=true and lucide-git-merge.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [L09-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f57d9f7647d9831a/L09-after-harness.mp4) |



## S04 — W×dark: real restricted pairing

Steps: pair separate disposable read-only clients with only orchestration:read and filesystem:read. Open source side pane while target pending. Hover for ~1.4 s to show the connection tooltip; physically click measured button centre, press real Enter, and retain ~1.2 s. After remains aria-pressed=true, aria-disabled=true; before remains disabled with no selected styling. Target pending remains unchanged. Final pair shows the connection tooltip before physical click/Enter. Date was restored to real time on both sides so its delay could complete.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [S04-before-readonly-final.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d6c05e7e1b9d4320/S04-before-readonly-final.mp4) | [S04-after-readonly-final.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/46d02982aae79d3f/S04-after-readonly-final.mp4) |



## M01 — I×light: selected/unselected native styling, PARTIAL still request

Steps: after starts selected; tap cancel ~3 s, remain fork, reopen Git menu ~8 s unselected. Before starts ordinary merge item, tap ~3 s repeats merge and navigates parent. This supplies missing light styling in real native recordings. Exact static before/after pairs and diff images remain excluded because normalized charging icon failed 500 ms identity checks.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [M01-before-ios-light-final.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/04d5925a68f73671/M01-before-ios-light-final.mp4) | [M01-after-ios-light.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d25ef93179e9d758/M01-after-ios-light.mp4) |



## M04-finish — A×light: real run completion continuation

Steps: start from previously accepted cancelled active-source fixture. Menu begins without merge item. Close it around ~10 s; resolve the real file-change approval ~14 s; Replay run finishes. Reopen ~30 s: ordinary unselected merge item returns. Retain accepted original selected/cancel/absent sequence; this is its missing finish supplement.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [M04-finish-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/aac73b94095511ee/M04-finish-after.mp4) |



## M06 — A×dark: labelled native command boundary

Initial selected item; injected client command defect on first tap ~6 s. Runtime feedback appears; reopen ~22 s and selected item remains. No parent navigation. Fault mechanism disclosure: matches.AsyncResult is undefined in the injected error branch, so it throws TypeError before RPC. This proves client command-defect feedback/retry, not a genuine server refusal. Hold branch invokes original real RPC after release. Final encoded frame was inspected separately because Android omits unchanged tail frames.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [M05-M06-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/04324df06e3d6823/M05-M06-after-harness.mp4) |

[Native dispatch transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5c87df18dc8a6a90/M05-M06-transcript.json).

## M05 — A×dark: labelled native command boundary

Hold retry after ~29 s; tap ~32 s, reopen ~39 s and tap again ~44 s. Transcript count remains one held RPC dispatch. Release ~50 s; reopen ~54 s ordinary unselected, staying source and without remerge. Fault mechanism disclosure: matches.AsyncResult is undefined in the injected error branch, so it throws TypeError before RPC. This proves client command-defect feedback/retry, not a genuine server refusal. Hold branch invokes original real RPC after release. Final encoded frame was inspected separately because Android omits unchanged tail frames.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| n/a: no cancel command before this PR | [M05-M06-after-harness.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/04324df06e3d6823/M05-M06-after-harness.mp4) |

[Native dispatch transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5c87df18dc8a6a90/M05-M06-transcript.json).

## M08 — A×dark: real keyboard-preserving overlay

Steps: source pending, focus composer to show actual software keyboard, open Git menu. At ~2 s tap source merge action; reopen ~8 s. After cancel leaves source and keyboard visible, and reopened overlay is unselected. Before repeats merge, navigates parent and dismisses keyboard. Same seeded ios-fork content; OS recording times differ.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [M08-before-android-dark.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f08fa2a1f6801cab/M08-before-android-dark.mp4) | [M08-after-android-dark.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5fbfd867c5d8c6e7/M08-after-android-dark.mp4) |

Additional after light keyboard-overlay flow: [M08-after.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4cbc0bd2376bf95a/M08-after.mp4).

## M07 / X01-native — A×dark: H hydration followed by real cross-client consumption

Steps: target never selected in this native client; real source newer run is awaiting approval. Before direct source navigation, real target atom is confirmed null. Title visibly labels HARNESS. Recorded opening menu has no item (~0–8 s); release real target subscription around ~10–16 s and selected item appears (~16 s sample). Dismiss menu ~24 s; web client sends target next message around ~30 s; reopen native menu at final ~41 s and item is absent while source still waits for approval. Target message is an actual command, not a projection write.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [M07-X01-before-native-final.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6d028fcf5597e571/M07-X01-before-native-final.mp4) | [M07-X01-after-hydration-harness-final.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/28c7ba3b5ecd7d34/M07-X01-after-hydration-harness-final.mp4) |

[Receipt and target-state transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d5620fb228cb6a68/native-target-consumption-transcript.json): merge receipt sequence 763; target next-send receipt sequence 813; target transfer consumed; source still running. Existing accepted X01 web/late-cancel proof is retained. Full cross-client reconnect/environment-switch row X03 is not established.

## M09 — Pixel Tablet×dark Files wrapper, PARTIAL

Steps: ios-fork is selected in the thread pane with Files inspector alongside. Initial menu selected; tap cancel ~2 s; remain on source with inspector. Reopen around ~8–12 s; final encoded frame is ordinary unselected merge item. Full Android split layout is visible; screenshot/encoded final frame inspected. Worker disconnect prevented baseline; later lease expired. iOS split layout was not reached.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| Not captured: Android baseline worker disconnected; iOS split not attempted within remaining budget. | [M09-after-tablet-matched-dark.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/13c281fd59f18292/M09-after-tablet-matched-dark.mp4) |



## What else changed and limits

- The L07 light static pair changes only the selected control background/glyph; both threshold diffs show this region alone.
- Interactive baseline and after histories evolve differently because baseline repeats merge and after cancels. Navigation, pending/superseded/consumed badges, source-vs-parent content and resulting toasts are expected consequences, not static pixel comparisons.
- Shared fixtures were reused across supplemental interactions. Some web parent timelines and sidebar order differ because of additional real sends/consumption. These videos demonstrate interaction branches; they are not claimed as identical static fixtures. Long-title light stills use their own matched fresh fixture.
- Native OS clocks, battery/status icons, real server timestamps, and elapsed Replay duration labels remain live. Newly visible later replies/approval cards come from the real run milestone being demonstrated. Native menu insertion/removal changes popup height and row wrapping. Keyboard dismissal differs when baseline merge navigates parent.
- Labelled H markers and source titles identify controlled faults, holds or selectors. Graph-shell omission intentionally changes parent status to Unknown. Missing-shell H intentionally changes fallback title/body. No H capture is represented as a real server failure or deletion capability.
- M06 is a client TypeError injection, not a server authorization/method rejection. Repeated-click web proof does not credit synthetic keydown as real Enter. Exact native still comparisons are withheld rather than accepting unstable frames.
- No CI, reviews, automated repository tests, Electron-only path, relay/tunnel transport, reconnect retention, environment-switch isolation, large-font scaling, or iOS tablet split is certified by this packet. No code was changed, so no code checks were run.

## Scripts and verification

All scripts are synthetic capture material, outside the worktree. They include actual seed commands, labelled harness boundaries, native command sequences, exact still stability/diff procedures and supplementary interaction notes. Runtime logs with tokens and failed/unsettled private captures are excluded from public uploads. Published media were inspected through decoded full-surface sequences and explicit ending frames, not accepted merely from script completion.

[build-evidence.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/94e28562711581b4/build-evidence.py)
[diff-stills.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/64a703c5cab92d46/diff-stills.py)
[finalize-evidence.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/8ae49f2523e71b04/finalize-evidence.py)
[fresh-native-active.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/22e07c541d227d46/fresh-native-active.js)
[init.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/23e947b2cb990992/init.js)
[interaction-driver-notes.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/068fd4f74e746ace/interaction-driver-notes.md)
[ios-active.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/00bc56c1403236d1/ios-active.js)
[mobile-seed.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d6a5ea930e668308/mobile-seed.js)
[native-env.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/edacde15982bcf4c/native-env.js)
[native-fault-sequence.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b4445dc7b0c9ec32/native-fault-sequence.py)
[native-harness.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d28a95d2f8fd0c17/native-harness.js)
[native-target-hydration.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0bff2f7e87b569d2/native-target-hydration.js)
[record-actions.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f579315a1f3eccf9/record-actions.py)
[seed-fallback.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/ad9b524b5b5a4487/seed-fallback.js)
[seed-fresh-native.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5a973d93bb372024/seed-fresh-native.js)
[seed-ios.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5d37d5da5cc1b9ca/seed-ios.js)
[seed-long-lineage.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/113bdcbb341a106e/seed-long-lineage.js)
[seed.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3f8171f12a8cfe2e/seed.js)
[still-native-no-status.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a462f97c6124f7d1/still-native-no-status.py)
[still-native.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/955c99e54d0f31ad/still-native.py)
[web-detach-harness.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/30d53403e2d308e1/web-detach-harness.js)
[web-fault-sequence.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/c37f5dfc9add6f67/web-fault-sequence.js)
[web-generic-parent-harness.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/dc90ea297d90adb8/web-generic-parent-harness.js)
[web-harness.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/10388730a4ccf850/web-harness.js)
[web-hold-sequence.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5c844322abe63745/web-hold-sequence.js)
[web-missing-shell.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0d0a916558a5c99c/web-missing-shell.js)
[web-target-hydration-sequence.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fee0fc5bc42f63de/web-target-hydration-sequence.js)
[web-target-hydration.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/43c7b45debf91a2d/web-target-hydration.js)

[build-evidence.py](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2941a84cbc5e7b51/build-evidence.py)
[interaction-driver-notes.md](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/e63ac05b8d124ff6/interaction-driver-notes.md)
[web-generic-parent-harness.js](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6ecde7ea7a6a3f2b/web-generic-parent-harness.js)

[M03-after-ios-active.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/642c138c7da39c87/M03-after-ios-active.mp4)

## Final qualifications and corrections

Final interpretation of the supplemental rows: S04 is complete: both final recordings include the actual connection tooltip before mouse/Enter, then retain the disabled state. The browser clock was restored to real time for these two recordings because freezing Date.now prevented the tooltip delay from finishing. This is a capture-clock adjustment, not a permission or state injection. The static L07 pair retained its frozen clock.

M03 is complete once the paired links below are read together: the baseline has no cancel item while the newer source run is active, then its normal merge item returns after Allow once and completion. The after client keeps selected cancellation available while active; cancel removes it; the same real approval/completion restores the unselected item. The native before runtime was checked to have cancelMergeBack absent and route environment equal the isolated baseline environment.

The M01 light recording satisfies the audit's missing native light styling; the brief's requested exact still pair remains unprovided. Treat M01 as partial against that stricter request, and retain existing accepted dark evidence.

X04 and M10: not captured: out of scope for this PR.

### M03 — I×dark paired native active-source recordings

After: 0 s selected while approval pending; cancel about 2 s; reopen absent about 8 s; dismiss about 12 s; Allow once about 15 s; finish and reopen unselected around 25 s. Before: no item while active at 0–4 s, so there is no cancel to tap; dismiss about 4 s; Allow once about 7 s; completion and ordinary merge item return about 15–18 s. Both remain on the source. OS clocks, newer-run send timestamps and elapsed waiting duration differ.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [M03-before-ios-active.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/482261a38d221e22/M03-before-ios-active.mp4) | [M03-after-ios-active.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/642c138c7da39c87/M03-after-ios-active.mp4) |


### Final M07/X01-native baseline and fixture differences

The native baseline sequence confirms the unavailable action on the merge-base. Before starts on the direct source fixture without target selected; opens Git menu near 0 s, which has no item because the newer source run is active. Target next-send is issued from the baseline web client around 20 s, consuming its pending transfer. Menu is dismissed/reopened; the final encoded frame (~40.5 s) still has no item while source waits. After's labelled hydration hold/release (~10–16 s) reveals selected cancellation; target next-send (~30 s) removes it by the ~41 s ending. The native source stays mounted in both recordings; neither native client selects its target. There is no cancel action on the baseline to fabricate.

The source Replay prompt differs: before “Check the active source,” after “Hold the source while its target loads.” Newer-run send time differs (before 5:08 PM, after 4:58 PM). Both use the same synthetic ordinal-1 completed / ordinal-2 running + file-change-approval structure, same title, same theme/device/window, and real target-owned pending→consumed transfer. This is qualified interaction evidence, not an identical static fixture pair. Baseline has no target-selected state to release; the after-only selector hold is labelled H.

| Before | After |
|---|---|
| [20c3dd3](https://git.cafe/pingdotgg/t3code/commit/20c3dd3689782bfcb1a04332086d08017028c718) | [c94024d](https://git.cafe/pingdotgg/t3code/commit/c94024dfa563d3a722d1f21ddcb24a9bca08419b) |
| [M07-X01-before-native-final.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/6d028fcf5597e571/M07-X01-before-native-final.mp4) | [M07-X01-after-hydration-harness-final.mp4](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/28c7ba3b5ecd7d34/M07-X01-after-hydration-harness-final.mp4) |

[native-before-consumption-transcript.json](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/cbc2bf33b6b9651d/native-before-consumption-transcript.json) records merge receipt 696, target send receipt 823, transfer consumed, source still running. The corresponding after receipt/state transcript is linked above.

## Cleanup and final verification

Owned after/before dev runners and both Metro runners stopped by their tracked sessions. No listeners remain on 5775, 5939, 13815, 13979, 8081 or 8082. Android and replacement iOS sessions returned Closed. Tablet lease had already ended by Simlock after worker disconnect; close then rejected without a lease, and its local proxy session was disconnected. No tablet was driven after expiration. All four owned preview tabs closed. The created before worktree was removed; the existing after checkout remains clean at the pinned head. No live install was modified. Scratch originals, excluded attempts, and private logs are retained locally. Available disk stayed above 230 GiB.

FRICTION:

tool | Intermittent pool daemon errors, tablet worker disconnect and lease expiration interrupted native work; retries after a minute were attempted | Stable workers and lease-health feedback would avoid lost capture time.

tool | Native semantic alert activation did not dismiss an unverified-foreground iOS deep-link prompt; screenshot-guided coordinate activation eventually succeeded | Use the documented AX-unavailable screenshot fallback immediately.

tool | Frozen Date.now prevented a delayed read-only tooltip from appearing | Freeze visible fixture time without freezing delay bookkeeping, or restore real Date for timed interaction recordings.

infra | Metro hit the host file-watcher limit during initial setup | CI=1 for headless Metro avoids interactive watcher pressure.

tool | CDP selection was shared across platform sessions; stale targets and several CLI flag mistakes caused retries | Read exact CLI help and select the intended live target before every native evaluation.

tool | Pillow was unavailable and native variable-frame-rate ending seeks sometimes yielded no PNG | ImageMagick for exact differences; decode/reverse the recording to inspect its true final frame.

AI-assisted; verified with the tests and tools listed above.
