### A6 — Android / Pi editor with empty prefill and zero options — recaptured on one lease
Evidence-only replacement for [#29420](https://git.cafe/pingdotgg/t3code/pull/29420): before [ac8dce0cc3](https://git.cafe/pingdotgg/t3code/commit/ac8dce0cc3abd8a337f2d806b5082a6834b9b81a); after [f16d8770bd](https://git.cafe/pingdotgg/t3code/commit/f16d8770bda7acff940ed18a3557c050aea6651e). Each side used its own server-created fresh synthetic database; head was repeated against a third fresh database. No source edits, commits, pushes or PR edits.
Environment: **one Pixel 9 lease on android-04**, Android 16/API 36, native **1080×2424 px**, logical 411.4286×923.4286, DPR 2.625, font scale 1, light, real English (US) Gboard with suggestions/autocorrection. **Status-bar height and top safe-area inset: 54.0952377 dp = 142 px**, confirmed independently by native StatusBar/SafeArea constants and a 142 px SystemUI AX rectangle on this lease; unchanged across all three runs. Native client fingerprint `971743703b097b8870b39ff842d130f50b115eeb`; agent-device 0.21.32; VM fs-01. Exact display-cutout overlay package and Gboard build are **unverified**: this agent-device has no shell/adb passthrough, and Settings did not expose the cutout entry. Native inset readings and geometry are recorded in [environment JSON](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/5646c3ea263791ab/environment.json). Server data clock is fixed at 2026-11-02T20:00:00Z; PNG status bar normalized to 9:41.
Steps and timestamps below are seconds from the record command; encoded video startup is approximate ±2 s. The recordings include generated spoken narration and an English subtitle track. Their native 1080×2424 video bitstreams are unchanged; raw videos and dispatch logs are retained.
| Step | Before | After | Head rerun |
|---|---:|---:|---:|
| Confirm empty Pi editor, `initialAnswer:""`, zero options; focus | 1.23s | 0.94s | 1.40s |
| Native Gboard taps: `helo` | 5.33s | 5.58s | 8.14s |
| Tap the observed `Hello` suggestion | 7.79s | 8.06s | 13.94s |
| Native Enter (multiline newline), then `cat` | 10.11s | 10.86s | 18.69s |
| Long press to open native selection toolbar | 28.62s | 15.60s | 27.48s |
| Select all; Gboard Delete; assert empty | 56.53s | 29.53s | 41.90s |
| Collapse and expand; assert empty again | 64.94s | 44.97s | 58.88s |
| Submit empty; verify provider receipt | 72.14s | 56.55s | 71.19s |
| Before | After | Diff0 | DiffT (>16) |
|---|---|---|---|
| [Before recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d3241adc3276d75a/before-A6-narrated.mp4)
[ac8dce0cc3](https://git.cafe/pingdotgg/t3code/commit/ac8dce0cc3abd8a337f2d806b5082a6834b9b81a) | [After recording](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/07e99d7d7d9c1634/after-A6-narrated.mp4)
[f16d8770bd](https://git.cafe/pingdotgg/t3code/commit/f16d8770bda7acff940ed18a3557c050aea6651e) | 
[ac8dce0cc3](https://git.cafe/pingdotgg/t3code/commit/ac8dce0cc3abd8a337f2d806b5082a6834b9b81a) → [f16d8770bd](https://git.cafe/pingdotgg/t3code/commit/f16d8770bda7acff940ed18a3557c050aea6651e) | 
[ac8dce0cc3](https://git.cafe/pingdotgg/t3code/commit/ac8dce0cc3abd8a337f2d806b5082a6834b9b81a) → [f16d8770bd](https://git.cafe/pingdotgg/t3code/commit/f16d8770bda7acff940ed18a3557c050aea6651e) |
All three Pi provider receipts are exactly `{"type":"extension_ui_response","id":"editor-1","value":""}`: [before](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/33b4ca1a9b9e8b72/before-A6-pi-receipt.jsonl), [after](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a35597409134bb4c/after-A6-pi-receipt.jsonl), [head rerun](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/86b78373e47a86dc/rerun-A6-pi-receipt.jsonl). The projections also confirm `initialAnswer:""`, `options:[]` and resolved requests. Empty submission is preserved as an empty string, without cancellation.
Final settlement: before screenshots [A](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/f67b82d2abf5f4e8/before-A6-final-a.png) / [B](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/47cb1f2b7ffa17d1/before-A6-final-b.png), after [A](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/543072b2b6b7f33e/after-A6-final-a.png) / [B](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fa4a3f654c21dac7/after-A6-final-b.png), rerun [A](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/28b927d7469c32dd/rerun-A6-final-a.png) / [B](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b6a38e565e6876cf/rerun-A6-final-b.png) are each pixel-identical. The driver requested a **500 ms wait between screenshot calls**, following the original capture method. Remote export/round-trip overhead made their command start gaps 4661.117 / 3022.890 / 3086.045 ms; **exactly 500 ms between native exposures is unverified**. See [settlement timing](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/bbea60df63a9003f/settlement.json). Before versus after final PNGs: **0 / 2,617,920 changed pixels (0%)**.
Reproducibility: [second head flow](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fc88f0f2327fd142/rerun-A6-narrated.mp4) on the **same lease**, third fresh database, produced **0 / 2,617,920 changed pixels (0%)** against the first head final still, with another empty-string Pi receipt.
Comparison with earlier evidence: the new head final differs from the [earlier delivered final](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/3ee394b4b551d88e/after-A6-v2-fixed-final.png) at **95,413 / 2,617,920 pixels (3.644611%)**. The earlier frame’s header is **79 px higher**, consistent with an **inferred 63 px / 24 dp top inset**, whereas the recapture uses the measured **142 px** inset. The entire **1080×420** right-hand header crop of the [auditor comparison](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/63574e4838b4b81f/rebuild-A6-header-comparison.png) matches the new head **exactly: 0 / 453,600 changed pixels**. Thus the recapture matches the auditor’s 142 px inset condition; the earlier delivery matches the smaller inferred 63 px condition. This supports an environment/inset explanation; it does **not** identify the exact overlay package. A full-frame auditor delta is **unverified**, because only its header-comparison image was supplied. [Pixel counts and comparison limits](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/4aa8ed491cdc6070/pixel-comparisons.json).
What else changed: the baseline’s intermediate edited value becomes `Geo \nCat` after the same native taps and observed suggestion target; head and rerun show `Hello \nCat`. All are then fully deleted. OS clock, signal indicators, transient status normalization, caret/pending-card pulses and keyboard predictions vary during recordings. The baseline includes a pause to inspect its native selection toolbar. Failed setup captures (Back navigating to Home, a keyboard-state read failure, and pairing/navigation readiness) are explicitly excluded and retained in the support archive. **No pixels differ between the accepted final surfaces.**
Diff0 is per-channel RGB absolute difference ×255, clamped to 255. DiffT marks PIL luminance of that absolute RGB difference >16 in magenta over a 50/50 image blend. No crop or alignment is used for full-frame comparisons. The auditor comparison is explicitly limited to its supplied header crop.
[Recapture runbook](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b0460c131d2bcf68/capture-runbook.md) · [updated driver (Enter alias)](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/315c45efe91d7bc4/driver.py) · [all supporting captures, raw videos, native trees, projections, scripts, timing logs and review sheets](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/fe7e2042b87b83e1/A6-supporting-artifacts.tar.gz). All PNG captures and whole-surface sampled recording sheets were opened before upload. Recording inspection used sampled frames; no frame-by-frame motion audit is claimed. Narration adds audio/subtitles and stream-copies the original video.
Cleanup verified: one lease released (lease lookup returns 404), all eight tracked seed/server/Metro processes exited, detached baseline checkout removed, head checkout clean. Credentials, pairing QR logs, databases and scratch HOME are excluded from uploads. Model/harness: gpt-6.1-sol through Codex.