Findings: no remaining D Light proof blocker for [#29263](https://git.cafe/pingdotgg/t3code/pull/29263). The previous independent-rebuild discrepancy is resolved by the supplied recapture and independently recomputed comparisons.

Row status: **D Light — SUFFICIENT**, first open and reopen. Reviewed exact head [3b020ab39b](https://git.cafe/pingdotgg/t3code/commit/3b020ab39b72d0a7be90f85515b41d22e5e7b87d), branch fix/issue-29201, against [47f2c8d29e](https://git.cafe/pingdotgg/t3code/commit/47f2c8d29e981c7723467efe137ee19e91df924e). Both the head parent and merge-base with origin/main equal that base. The reviewed checkout is clean. Local main points elsewhere; origin/main and the requested immutable SHAs were used for this proof audit.

Provenance and visual checks: the before/after column SHAs match those immutable revisions. Both primary takes are 1080×2424, density 2.625, system fontScale 1, explicit Light/T3 Code, app text 16 pt, root (0,0,1080,2424), Gboard (0,1541,1080,883), status142/nav63. Viewed all eight selected new full-device checkpoints, the two original/archive comparison inputs, all 20 supplied comparison/diff/crop sheets, and the distinct raw screenshot samples. Decoded all four raw recordings (733 video frames total); viewed every menu/header frame plus full-device frames throughout, and inspected native-resolution frames where contact-sheet text was ambiguous. The selected after states have complete Environment/Project rows, no ghosting, stale overlay, spinner/skeleton, clipped text, unintended blank region, or wrong theme/viewport. The only baseline clipping is the reported missing Project row.

The written recording narration matches first open, settled capture, outside dismissal, reopen, settled capture and final dismissal. Gboard remains visible. Command-completion timestamps are approximate offsets, as disclosed; they are not frame-exact transition measurements. The recordings retain live/normalized status-clock changes from the explicit screenshot option. No audio narration was claimed or needed; narration is the supplied written step tables. All eight selected stills have independent identical screenshot partners, and all eight published PNGs preserve the raw RGB pixels. All four published MP4s are byte-identical to their raw recordings.

Independently recomputed full-frame RGB results:

| Comparison | Pixels changed >0 | Pixels changed >8/255 | Bounding box |
|---|---:|---:|---|
| Before → after, first open | 38,696 | 17,904 | (378,445)–(1063,609) |
| Before → after, reopen | 38,696 | 17,904 | (378,445)–(1063,609) |
| First after → cold equal-inset after, first open | 262 | 259 | (248,188)–(253,244) |
| First after → cold equal-inset after, reopen | 262 | 259 | (248,188)–(253,244) |
| New android-03 first open → capture 2b delivered D | 262 | 259 | (248,109)–(253,165) |
| New android-03 reopen → capture 2b delivered D | 0 | 0 | none |

All eight supplied primary/equal-inset Diff0/DiffT masks exactly match recomputation using their actual pink highlight RGB (255,40,150). Primary differences are solely the lower menu boundary/shadow and revealed Project label/chevron; every other primary before/after pixel is identical. This matches the sole source change, replacing the inline keyboard-preserving Animated.View/FadeIn wrapper with View in AndroidAnchoredMenu.tsx. The equal-inset independent residual is solely the blinking search caret (maximum channel delta 228), with identical app/menu geometry on both opens.

Inset explanation: the separate native probes read status/cutout top142 on android-04 and status top63/displayCutout=null on android-03, nav63 on both. Native T3 constants agree: 54.0952377 dp ×2.625 ≈142 px versus 24 dp ×2.625=63 px. Each take's own raw snapshot also corroborates root/status/IME geometry. Search y167→88, Environment y346→267 and Project y472→393 all move by exactly79 px; x/width/height match. Translating the documented 705×340 menu crop by79 px leaves just two one-level pixels, zero above8. Gboard remains fixed and has 115,151 raster differences of at most5/255 across hosts; body centering accounts for the half-inset empty-state shift. Those cross-host diagnostics are not used as the primary PR before/after comparison.

The earlier audit report and audit-core archive downloaded successfully. Its independent D input differs from the new primary after only by the same262 caret pixels, and is pixel-identical to the new cold equal-inset D input. Its previously reported375,511 changed pixels against the original capture2b image are independently reproducible. Thus both original rendering environments are now reproduced and the prior finding is explained. The earlier snapshot's137 px notification-node rectangle was not a native WindowInsets measurement; the new native probes, app constants and exact app pixel identity substantiate the effective142 px top inset.

Scope: only D Light was re-checked; the other15 previously accepted rows were not re-audited. No new device rebuild was required. No source edits, pushes, comments, PR mutations, test servers, emulators, leases or worktrees were created. No CI checks or general code audit were performed. Completed within the25-minute budget. Derived audit calculations and visual-review sheets are preserved in the accompanying archive before removal of this step's scratch directory. Existing evidence, the scratch root and /home/ubuntu/wt/pr-29263 are preserved.

FRICTION: none

VERDICT: SUFFICIENT