# Independent proof audit, pass 2

PR [#22877](https://github.com/pingdotgg/t3code/pull/22877), base [e54e71d725](https://github.com/pingdotgg/t3code/commit/e54e71d725b6eef166ee9b85137c09e051cff342), head [e8a30afda4](https://github.com/pingdotgg/t3code/commit/e8a30afda42d34ad8f0627d40b864c260fab7f40).

Verdict: INSUFFICIENT. Zero of 17 matrix rows complete; H1 and C1 partial. H1 lacks the full before-stall recording and narrow/dark pairs. C1 lacks a matched hidden RED baseline: before RED completed before hiding, while BLUE timed out. Dark coverage is missing. H2–H8, N1–N2, C2–C4, W1, R1, D1 remain unverified, as listed in the final evidence index.

Independent rebuild at final head passed. Real Linux Electron, committed fixture, renderer 1440×900, native guest 539×820, DPR 1. Both RED and BLUE server-tool captures returned while B remained selected. Rendering stayed active beyond 10 seconds and suspended after the turn completed. Both returned images exactly match the delivered head PNGs: zero changed pixels at zero threshold. All four delivered diff pairs exactly match recomputed zero-threshold 16× and >8/255 masks. The disclosed control-bar, letterboxing, rasterization, chat, timing, and surrounding-state differences are consistent with source and images.

The independent rebuild strengthens C1 behavior proof; it does not replace missing matched matrix coverage. All owned runtime processes and browser connections stopped; ports 14806, 14807, 19291 clear; source/worktree unchanged.

Audit: GPT-6.1 Sol in Codex/T3 Code.