Addresses Apr 24 review of 58fa8c52. All 6 findings landed.
Bug #1 — git log -1 returns latest commit, not first (semantic mismatch
with "original author" comment):
Drop -1 flag, take last line of default-ordered log output (= oldest).
Fixes mis-credit on multi-commit PRs where a reviewer rebased/force-pushed.
Nit #2 — forward writer didn't pass merged_at:
Fetch merged_at in the prs SELECT, thread pr_merged_at through all 5
insert_contribution_event call sites. Keeps forward-emitted and backfilled
event timestamps on the same timeline after merge retries.
Nit #3 — legacy-counts fallback paths emit no events (parity gap):
git-author and prs.agent fallback paths now emit challenger/synthesizer
events via the TRAILER_EVENT_ROLE map when refined_type matches. Closes
the gap where external-contributor challenge/enrich PRs would accumulate
legacy counts but disappear from event-sourced leaderboards.
Nit #4 — migration v24 agent seed missing 'pipeline':
Added "pipeline" to the seed list. Plus new migration v25 with idempotent
corrective UPDATE so existing envs (where v24 already ran) pick up the
fix on restart without requiring manual SQL. Verified on VPS state:
pipeline row was kind='person', will flip to 'agent' on redeploy.
Nit #5 — backfill summary prints originator attempted=0 in wrong pass:
Split the "=== Summary ===" header into "=== PR-level events ===" and
"=== Claim-level originator pass ===" with originator counts in the
right block. Operator-facing cosmetic.
Refactor #6 — AGENT_BRANCH_PREFIXES duplicated in 2 sites:
Extracted to lib/attribution.py as single source of truth. contributor.py
imports it. backfill-events.py keeps its local copy (runs standalone
without pipeline package import) with a sync-reference comment.
No behavioral drift for the common case. Backfill re-runs cleanly against
existing forward-written events (UNIQUE-index idempotency).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Introduces contribution_events table + non-breaking double-write. Schema
lands today, forward traffic writes events alongside existing count upserts,
backfill script replays history. Phase B will add leaderboard API reading
from events; Phase C switches Argus dashboard over.
## Schema v24 (lib/db.py)
- contribution_events: one row per credit-earning event
(id, handle, kind, role, weight, pr_number, claim_path, domain, channel, timestamp)
Partial UNIQUE indexes handle SQLite's NULL != NULL semantics:
idx_ce_unique_claim on (handle, role, pr_number, claim_path) WHERE claim_path NOT NULL
idx_ce_unique_pr on (handle, role, pr_number) WHERE claim_path IS NULL
PR-level events (evaluator, author, challenger, synthesizer) dedup on 3-tuple.
Per-claim events (originator) dedup on 4-tuple. Idempotent on replay.
- contributor_aliases: canonical handle mapping
Seeded: @thesensatore → thesensatore, cameron → cameron-s1
- contributors.kind TEXT DEFAULT 'person'
Migration seeds 'agent' for known Pentagon agent handles.
## Role model (confirmed by Cory Apr 24)
Weights: author 0.30, challenger 0.25, synthesizer 0.20, originator 0.15, evaluator 0.05
- author: human who submitted the PR (curation + submission work)
- originator: person who authored the underlying content (rewards external creators)
- challenger: agent/person who brought a productive disagreement
- synthesizer: cross-domain work (enrichments, research sessions)
- evaluator: reviewer who approved (Leo + domain agent)
Humans-are-always-author: agents credit is capped at evaluator/synthesizer/
challenger. Pentagon agents classify as kind='agent' and surface in the
agent-view leaderboard, not the default person view.
## Writer (lib/contributor.py)
- New insert_contribution_event(): idempotent INSERT OR IGNORE with alias
normalization + kind classification. Falls back silently on pre-v24 DBs.
- record_contributor_attribution double-writes alongside existing
upsert_contributor calls. Zero risk to current dashboard.
- Author event: emitted once per PR from prs.submitted_by → git author →
agent-branch-prefix.
- Originator events: emitted per claim from frontmatter sourcer, skipping
when sourcer == author (avoids self-credit double-count).
- Evaluator events: Leo (always when leo_verdict='approve') + domain_agent
(when domain_verdict='approve' and not Leo).
- Challenger/Synthesizer: emitted from Pentagon-Agent trailer on
agent-owned branches (theseus/*, rio/*, etc.) based on commit_type.
Pipeline-owned branches (extract/*, reweave/*) get no trailer-based event —
infrastructure work isn't contribution credit.
## Helpers (lib/attribution.py)
- normalize_handle(raw, conn=None): lowercase + strip @ + alias lookup
- classify_kind(handle): returns 'agent' for PENTAGON_AGENTS, else 'person'
Intentionally narrow. Orgs get classified by operator review, not heuristics.
## Backfill (scripts/backfill-events.py)
Replays all merged PRs into events. Idempotent (safe to re-run). Emits:
- PR-level: author, evaluator, challenger, synthesizer
- Per-claim: originator (walks knowledge tree, matches via description titles)
Known limitation: post-merge PR branches are deleted from Forgejo, so we
can't diff them for granular per-claim events. Claim→PR mapping uses
prs.description (pipe-separated titles). Misses some edge cases but
recovers the bulk of historical originator credit. Forward traffic gets
clean per-claim events via the normal record_contributor_attribution path.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Prevents Apr 22 runaway-damage pattern (44 open PRs manually bulk-closed)
where a source extracted 20+ times before the cooldown gate landed, each
leaving an orphan 'open' PR after eval correctly rejected as near-duplicate.
Gate fires in dispose_rejected_pr before attempt-count branches:
all_issues == ["near_duplicate"] (exact match — compound carries signal)
AND sibling PR exists with same source_path in status='merged'
AND diff contains "new file mode" (not enrichment-only)
→ close on Forgejo + DB with audit, post explanation comment.
Ganymede review — 5 must-fix/warnings applied + 1 must-add:
- Exact match on single-issue near_duplicate (compound rejections preserved)
- Enrichment guard via diff scan (eval_parse regex can flag enrichment prose)
- 10s timeout on get_pr_diff — conservative fallback on Forgejo wedge
- Forgejo comment with canned explanation (best-effort, try/except)
- Partial index idx_prs_source_path + migration v23
- Explicit p1.source_path IS NOT NULL in WHERE
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Ganymede review findings:
1. source_channel was missing from CREATE TABLE (fresh installs wouldn't have it)
2. Default fallback changed from 'telegram' to 'unknown' — unknown prefixes
are genuinely unknown, not telegram
3. Cross-reference comments added between BRANCH_PREFIX_MAP and _CHANNEL_MAP
Also wires classify_source_channel into merge.py PR discovery INSERT.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Enables GitHub↔Forgejo PR linking for the contributor pipeline.
Mirror script will store GitHub PR number when creating Forgejo PRs,
allowing back-sync of eval feedback and merge/close status.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
db.py: migration v20 adds conflict_rebase_attempts, merge_failures,
merge_cycled columns (already exist on VPS via manual migration, missing
from code — any future DB rebuild would break retry mechanism).
merge.py: replace retry-with-backoff on config.lock with asyncio.Lock
(_bare_repo_lock) around all worktree add/remove calls. Prevents
contention instead of retrying it. Applied to both _cherry_pick_onto_main
and _merge_reweave_pr.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Atomic extract-and-connect (lib/connect.py):
- After extraction writes claim files, each new claim is embedded via
OpenRouter, searched against Qdrant, and top-5 neighbors (cosine > 0.55)
are added as `related` edges in the claim's frontmatter
- Edges written on NEW claim only — avoids merge conflicts
- Cross-domain connections enabled, non-fatal on Qdrant failure
- Wired into openrouter-extract-v2.py post-extraction step
Stale PR monitor (lib/stale_pr.py):
- Every watchdog cycle checks open extract/* PRs
- If open >30 min AND 0 claim files → auto-close with comment
- After 2 stale closures → marks source as extraction_failed
- Wired into watchdog.py as check #6
Response audit system:
- response_audit table (migration v8), persistent audit conn in bot.py
- 90-day retention cleanup, tool_calls JSON column
- Confidence tag stripping, systemd ReadWritePaths for pipeline.db
Supporting infrastructure:
- reweave.py: nightly edge reconnection for orphan claims
- reconcile-sources.py: source status reconciliation
- backfill-domains.py: domain classification backfill
- ops/reconcile-source-status.sh: operational reconciliation script
- Attribution improvements, post-extract enrichments, merge improvements
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Schema migration v3: adds eval_attempts (INTEGER) and eval_issues (TEXT/JSON)
columns to prs table.
Retry budget logic (Ganymede-approved design):
- Increment eval_attempts on each evaluate_pr() call
- Hard cap: eval_attempts >= 3 → terminal (close PR, tag source needs_human)
- Attempt 1: normal — back to open, wait for fix
- Attempt 2: classify issues as mechanical/substantive
- Mechanical only (schema, wiki links, dedup): keep open for one more try
- Substantive (factual, confidence, scope, title): close PR, requeue source
- Issue tags parsed from reviewer comments, stored in eval_issues column
- SHA-based reset: new commits on PR branch → eval_attempts=0, verdicts reset
- Post-migration stagger: LIMIT 5 for first batch to avoid OpenRouter spike
- Cost recording updated: domain review → OpenRouter, Leo → tier-dependent
Stops the 32-PR infinite loop burning ~$0.03/cycle with no terminal state.
Pentagon-Agent: Leo <294C3CA1-0205-4668-82FA-B984D54F48AD>
