Log t88 phase 2b: GPU lighting checks, profile, and the gate mistake

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/agent-log.md

@@ -4,6 +4,22 @@ What agents have done, newest first. Add an entry for every piece of work: the d
 note or request it was for, what changed, the files touched, the tests run, and anything left
 open or not verified. Keep each entry short; the git history has the details.
 
+## 2026-10-27: t88 phase 2b, lighting on the GPU (Codex + Claude Code)
+
+- `src/render/gpu/light.js`: light sources per light cell (transmittance, sky from `roofY`, heat,
+  bot lamp), propagation as a left/up then right/down relaxation on float textures (the same
+  light paths as the CPU sweep), then packing. The WebGPU path no longer calls `computeLight`;
+  the upload rectangle widens by the light margin. Fixed CPU light buffers keeping stale sizes
+  after a renderer switch. Scene `tools/scenes/t88-light.js` (lamp, opening, magma, fire).
+- Checked: light scenes legacy vs WebGPU, ImageMagick MAE 16-27/65535 (<0.1/255), identical by
+  eye; `test/t_gpu.py` light parity in full and fast modes. Profile (factory, pixel): drawWorld at
+  zoom 2 24.55 (baseline) -> 9.98 (2a) -> 6.79 ms, CPU light 0, uploads now the biggest part
+  (4.87 ms); real loop zoom 2 CPU p50/p95 31.7/35.3 -> 10.25/11.48 ms, a steady 60 fps.
+- Commit 31e3005 went in after a full-suite run where `t_gpu.py` timed out once (a 30 s page wait
+  under load): the commit sat after a heredoc, outside the test gate. The suite was re-run and
+  passed in full, and `t_gpu.py` passes alone. Codex hit its usage limit before the final
+  checks; Claude Code ran them.
+
 ## 2026-10-27: t88 phase 2a, WebGPU pixel look (Claude Code + Codex)
 
 - New `src/render/gpu/` (device, textures, colour compute, glow blur, draw) renders the pixel look