bmottin/claude_mac
1
0
Fork 0
forked from bchanot/claude
Code Pull Requests Activity

chore(darwin): T3 baseline — 54 rows, mean 83.4, 13 candidates <80

Browse Source
This commit is contained in:
Bastien Chanot
2026-08-26 11:15:03 +02:00
parent afec610dc8
commit af002ba235
1 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.claude/tasks/TODO.md
+5 -2
View File
@@ -13,8 +13,11 @@ paired same-judge majority, absolute scores triage-only.
EXCLUDED — machine-owned ctx7 (BDR-053, gitignored) → 31 systems. EXCLUDED — machine-owned ctx7 (BDR-053, gitignored) → 31 systems.
- [x] T2 Phase 0.5 gate PASSED: reuse prompts as-is; dim8 full_test on - [x] T2 Phase 0.5 gate PASSED: reuse prompts as-is; dim8 full_test on
candidates only (baseline dry_run); Phase 2 set = ALL units <80. candidates only (baseline dry_run); Phase 2 set = ALL units <80.
- [ ] T3 Phase 1 baseline: batched blind judges, 9-dim, 55 rows in - [x] T3 Phase 1 baseline DONE: 7 blind judges, 54 rows (31 skills + 23
results.tsv; scorecard. agents), mean 83.4, 13 units <80, ~25 verified findings (hotfix
destructive restore, onboard/onboarder contract, init-project
allowed-tools, skills-perso 8/32 detection...).
- [ ] T4 Phase 1 gate: scorecard checkpoint, user picks optimization set. - [ ] T4 Phase 1 gate: scorecard checkpoint, user picks optimization set.
- [ ] T5 Phase 2: per-unit loops (1 dim/round, weighted-gap diagnosis, - [ ] T5 Phase 2: per-unit loops (1 dim/round, weighted-gap diagnosis,
paired 3-judge majority keep/revert, HL-4 stop), human checkpoint paired 3-judge majority keep/revert, HL-4 stop), human checkpoint