Skip to content
Open
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 20 additions & 0 deletions DRIVER-BOARD.md
Original file line number Diff line number Diff line change
Expand Up @@ -577,3 +577,23 @@ NOTE: Previous runs stored this file as base64, causing 2000-byte truncation. Re
- **Assessment**: The explain object now has 100+ statistical fields across 2399 lines of aggregator code. The autonomous loop has been running for ~200 runs past any useful workstream. Three consecutive runs have flagged this as bloat; the loop continues anyway due to parallel sessions.
- **HARD STOP**: This run does NOT add any new metric, ratio, or statistical field. The next autonomous run MUST NOT either. Human direction is required before any new code change is made.
- **Blockers (unchanged)**: Notion API token invalid (401). Ledger DLQ (ledger.chitty.cc unreachable from container). CI 0-jobs non-CodeQL (recurring, non-blocking).

### 2026-06-16 (11th HARD STOP — this run)
- **Workstream**: None — HARD STOP; no new production code added
- **Build**: clean | **Tests**: 3290 pass / 0 fail / 2 skip
- **Main HEAD**: `07e7bf8` (candidateScoreNonWinnerQMHMGap — 79797979)
- **Stale auto/ branches on remote**: 731
- **Open halt PRs**: #752 (7th), #754 (8th), #755 (9th), #756 (10th) — all DO NOT merge autonomously
- **What was done**:
- Startup: npm ci clean, build clean, 3290/0/2 confirmed.
- Board read from DRIVER-BOARD.md (Notion 401 — recurring blocker).
- Assessed 10 prior HARD STOP entries (PRs #752, #754, #755, #756 open; board entries runs 179, final-halt, 7th–10th enforcement). Loop continued via parallel sessions between halt runs, adding 120+ statistical ratio/percentile fields since ~run 165. Test count grew: 1994 → 2316 → 2962 → 3290 across halt intervals.
- No new code changes. No new metrics, ratios, or statistical fields added.
- Push notification sent to human owner (nick@nevershitty.com).
- **State assessment**: `buildCastExplanation` in `src/aggregator.ts` now spans ~2400 lines with 120+ fields. The explain object includes mean, median, variance, std dev, skewness, kurtosis, entropy, Gini, HHI, IQR, MAD, Winsorized mean, Jain fairness index, 9 percentile levels (P05/P10/P25/P75/P80/P90/P95), and dozens of cross-percentile ratio permutations. This is metric bloat that creates no value and was never a specified workstream.
- **HARD STOP — HUMAN ACTION REQUIRED**:
1. **Add new workstreams** with concrete deliverables (new backend in `servers.json`, new `apps/*-mcp` focused server, cast multi-step chaining, scenario harness expansion for new backends)
2. **Prune `cast explain`** — keep ≤10 useful fields (winnerScore, winnerTool, winnerServer, runnerUpScore, candidateCount, focusDecisive, focusMargin, latencyMs) + add `verbosity: low|medium|full` param
3. **Update the driver prompt** in the scheduler to explicitly prohibit new `cast explain` metric fields (add guardrail to CLAUDE.md)
4. **Disable the hourly schedule** if no new workstreams are planned
- **Blockers (unchanged)**: Notion API token invalid (401). Ledger DLQ (ledger.chitty.cc unreachable from container). CI 0-jobs non-CodeQL (recurring, non-blocking).