erik/acdream
1
0
Fork 0
Code Issues Pull requests Projects Releases 6 Packages Wiki Activity Actions

docs(render): ledger — E1 re-measured same-session: every stationary destination within 1.14x, chunk 3 back to CLOSED; file #463 (soak turn-sample artifact)

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
Erik 2026-09-03 19:09:03 +02:00
parent a614c8fd4e
commit 6c7512de15
2 changed files with 22 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

21
docs/ISSUES.md
View file

@ -39,6 +39,27 @@ confirmed closed by the owner, 11 need a focused live gate, and 43 are safe to
remain closed. See
[`docs/research/2026-08-28-owner-closed-issue-validity-audit.md`](research/2026-08-28-owner-closed-issue-validity-audit.md).
## #463 — Soak `turn` samples can report a 0.1 ms GPU p50 (the turn window catches non-world frames)
**Status:** OPEN — found 2026-09-03 by the Campaign OVERHAUL v2 S3 review's perf re-measure.
**Severity:** LOW (measurement only; the stationary samples and every p95 are sane)
**Component:** `tools/run-connected-r6-soak.ps1` / the Cli soak's per-destination `turn` sample
**Symptom:** in `.claude/worktrees/s2-final-ab/logs/connected-r6-soak-20260903-185904.report.json`
(S2-final binary) the `turn` phase reports `GpuP50Ms = 0.10` at Caul, Holtburg and Caul plateau
while the same destinations' `stationary` p50 is 5.8 / 2.5 / 5.8 ms and the `turn` p95 is 10.5 /
3.0 / 8.0 ms; the candidate run nine minutes later (`logs/connected-r6-soak-20260903-185034.report.json`,
tip `a614c8fd4`) reads 0.1 / 2.3 / 4.3 for the same rows. A median of 0.1 ms under a p95 of 8–10 ms
means more than half the frames in that turn window drew no world (portal space, a load hitch, or
the reveal's wait cue), so the per-destination ratio a reviewer computes from `turn` p50s (23× and
43× here) is noise. The chunk-3 perf gate in the campaign ledger was first misjudged on the other
phase for the same reason (only one phase quoted).
**Fix direction:** either start the turn window only after the destination's reveal has completed
AND a fixed number of world frames have presented, or report the turn sample's world-frame count so
a 0.1 ms median is flagged; the stationary sample already waits. Until then, judge perf gates on the
`stationary` phase and treat a turn p50 under 0.5 ms with a multi-ms p95 as invalid.
## #462 — Login reveal into an interior cell stays `ready=True materialized=False` for 180 s (foundry `0xA9B40176`)
**Status:** OPEN — found 2026-09-03 by the Campaign OVERHAUL v2 G3 self-gate part C (`logs/selfgate-20260903-165745-g3c`, Release at `2fbfdf18a`).

2
docs/plans/2026-09-01-campaign-overhaul-world-solidity.md
View file

File diff suppressed because one or more lines are too long