bchanot/bchanot-cv
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

chore(memory): capitalize — LRN-004, EVAL-001, BDR-006 update note, journal 2026-07-06

Browse Source
This commit is contained in:
Bastien Chanot
2026-07-06 01:06:54 +02:00
parent dd8c327162
commit c8c72c24aa
4 changed files with 26 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.claude/memory/evals.md
+9
View File
@@ -21,6 +21,15 @@ rules:
| ID | Date | Output | Action |
|----|------|--------|--------|
| EVAL-001 | 2026-07-06 | /tour run 2026-07-05-3 (3 it., converged) + residual closure | keep |
## EVAL-001 — /tour run 2026-07-05-3 + residual closure pass
- **Date**: 2026-07-06
- **Output**: 3-iteration tour (security/clean/reconcile/doc, converged at bound) + closure of all 10 residuals on owner GO. Commits `1aa97f0`/`613bfc0`/`2f5e51a` + follow-up.
- **Method**: oracle-based — semgrep ×3 (deterministic PASS), PDF render-hash (LRN-003) for behavior-preserving proofs, docker oracles (build, nginx -t, header/dotfile/gzip/healthcheck curls), headless-browser screenshots 375+1440 (index font trim), brace counts, CSP-hash pinned==computed.
- **Anomalies**: (1) cso add-on caught a HIGH (base-image CVE) two same-day semgrep-only tours missed — gstack was OFF then → LRN-004. (2) Fresh clean sweeps surfaced new info-tier nits each iteration (N1–N4 at it2) — convergence needed explicit reporting threshold in it3 prompt; bound of 3 did its job. (3) Session limit killed both it3 agents mid-flight — SendMessage transcript-resume recovered both, zero re-audit gap.
- **Action**: keep
<!-- Append entries below. Template: