fix(attribution): credit research-session sources to agents, not m3taversal #7

Merged

m3taversal merged 5 commits from ship/research-attribution-fix into main

2026-04-27 11:59:55 +00:00

m3taversal commented

2026-04-27 11:39:03 +00:00

Owner

Two-part fix. Forward fix in research-session.sh frontmatter template; backfill script for last 30 days already applied to production (304 PRs).

Pre-backfill DB snapshot: pipeline.db.bak-pre-research-attribution on VPS.

Two-part fix. Forward fix in research-session.sh frontmatter template; backfill script for last 30 days already applied to production (304 PRs). Pre-backfill DB snapshot: `pipeline.db.bak-pre-research-attribution` on VPS.

m3taversal added 3 commits 2026-04-27 11:39:03 +00:00

fix(attribution): gate writer on publishers table (regression prevention) d0fb4c96e3

Schema v26 (commit 3fe524d) split orgs/citations from contributors into
the publishers table. Without a writer-side gate, every merged PR with
`sourcer: cnbc` (or similar) re-creates CNBC as a contributor and
undoes the v26 classifier cleanup. Once normal pipeline traffic resumes,
the contributors table re-pollutes within hours.

Fix: belt-and-suspenders gate at both writer surfaces.

1. `lib/attribution.py::is_publisher_handle(handle, conn)` — returns
   publisher.id if handle exists in publishers.name, else None. Falls
   back gracefully on pre-v26 DBs (no publishers table → returns None →
   writer behaves like before, no regression).

2. `lib/contributor.py::insert_contribution_event` — checks
   is_publisher_handle on canonical handle before INSERT. If it's a
   publisher, debug-log + return False. Prevents originator events for
   CNBC/SpaceNews/etc.

3. `lib/contributor.py::upsert_contributor` — same gate at top. Prevents
   the contributors table from re-acquiring publisher rows.

Verified end-to-end against live VPS DB snapshot:
  - CNBC originator event: blocked (insert returns False)
  - CNBC contributors row: blocked (no row created)
  - alexastrum, thesensatore, newhandle_xyz: pass through unchanged
  - is_publisher_handle handles case-insensitive lookup correctly
    (CNBC and cnbc both match publisher_id=3)

Pre-deploy event count was 3705. Post-classifier cleanup: 3623 (82 org
events purged). Going forward, no new org events accumulate.

Branch 2 of the schema-v26 rollout. Branch 3 (auto-create at tier='cited',
extract.py sources.publisher_id wiring) is separate scope and not required
for regression prevention.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

fix(attribution): narrow exception + document gate asymmetry (Ganymede review) dea1b02aa6

Two follow-up fixes from Ganymede's review of d0fb4c9:

1. is_publisher_handle: narrow `except Exception` to sqlite3.OperationalError.
   Pre-v26 DB fallback only needs to catch the "table doesn't exist" case;
   broader exceptions (programming errors, locks, corruption) should propagate.

2. upsert_contributor gate: add comment documenting the alias-resolution
   asymmetry between insert_contribution_event (alias-resolved via
   normalize_handle) and upsert_contributor (bare lower+lstrip-@). Today this
   is fine because the v26 classifier produced one publisher row per canonical
   handle. Branch 3 will normalize alias→canonical at writer entry points,
   tightening this gate transparently.

Unit tests for the gates (positive + negative + alias resolution) deferred to
Branch 3 alongside the auto-create flow tests.

Smoke-tested:
  - pre-v26 fallback (no publishers table) → None (correct)
  - case-insensitive match (CNBC → id=1) → correct
  - @ prefix strip (@cnbc → id=1) → correct
  - non-publisher handle (alexastrum) → None (correct)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

fix(attribution): credit research-session sources to agents, not m3taversal

CI / lint-and-test (pull_request) Waiting to run

Details

2d332c66d4

Two-part fix for a bug where every claim extracted from agent overnight
research sessions was being credited to m3taversal in contribution_events
(visible in the activity feed as "@m3taversal" on agent-derived claims).

Forward fix (research/research-session.sh):
The frontmatter template the agent prompt instructs Claude to use now
includes `proposed_by: ${AGENT}` and `intake_tier: research-task`. With
those fields present, extract.py path 1 (line 687) takes precedence and
sets prs.submitted_by to the agent handle, which then propagates into
contribution_events as a kind='agent' author event for the agent.

Without the fields, extract.py fell through to the default branch on
line 695 and set submitted_by='@m3taversal'.

Backfill (scripts/backfill-research-session-attribution.py):
Identifies research-session-derived PRs by finding teleo-codex commits
matching `^<agent>: research session YYYY-MM-DD —`, listing the
inbox/queue/*.md files added in each commit's diff, and matching those
filename basenames against prs.source_path. Only PRs currently
submitted_by='@m3taversal' AND merged within the configurable window
are touched. Default --dry-run; --apply to commit.

For each match the script:
  1. UPDATE prs SET submitted_by = '<agent> (self-directed)'
  2. INSERT OR IGNORE the agent author event (kind='agent', weight=0.30)
     with the original PR's domain, channel, merged_at preserved
  3. DELETE the misattributed m3taversal author event

Applied 30-day backfill on VPS:
  - 304 PRs re-attributed (rio 74, clay 70, astra 53, vida 48,
    theseus 30, leo 29)
  - 297 m3taversal author events deleted, 304 agent author events
    inserted (delta of 7 = pre-v24 PRs that never had m3ta events
    in the first place; we still create the new agent event)
  - m3taversal author count: 1368 → 1071 (−22%)
  - Pre-backfill DB snapshot: pipeline.db.bak-pre-research-attribution

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

rio added 1 commit 2026-04-27 11:50:20 +00:00

test(attribution): prove research-backfill replay is idempotent

CI / lint-and-test (pull_request) Waiting to run

Details

319e03e2c6

Five tests against the real contribution_events schema (lib/db.py:181-209):
- pr-level dedup with NULL claim_path via idx_ce_unique_pr partial index
- per-claim dedup with non-NULL claim_path via idx_ce_unique_claim partial index
- pr-level and per-claim events coexist on the same pr_number
- backfill (INSERT correct + DELETE wrong) is a true no-op on replay
- replay against already-backfilled state preserves unrelated events

Schema case identified: case 2 with partial-index split solution already in
place. Two partial UNIQUE indexes target disjoint row sets (claim_path IS NULL
vs IS NOT NULL), bypassing SQLite's NULL-not-equal-NULL UNIQUE quirk.

Production replay verified: re-running backfill --apply against the live DB
returns "misattributed PRs found: 0" because the first-run UPDATE flipped the
WHERE predicate. Total contribution_events count: 3839 → 3839.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

rio added 1 commit 2026-04-27 11:53:57 +00:00

fix(attribution): unify research-session format on "(self-directed)" suffix

CI / lint-and-test (pull_request) Waiting to run

Details

6aff03ff56

Resolves the format inconsistency between the forward fix and the 304-row
backfill. Both halves now produce prs.submitted_by = "rio (self-directed)":

- research-session.sh: drop proposed_by from the frontmatter template.
  extract.py path 1 (proposed_by-driven) no longer fires; path 2 fires
  instead and constructs f"{agent} (self-directed)" — matches backfill.

- attribution.py: normalize_handle now strips "(self-directed)" suffix
  immediately after lowercase+@-strip, before alias lookup. Closes the
  phantom-person-event class on any future replay through
  record_contributor_attribution. Round-trips through alias rules keyed
  on bare agent names.

Test (5 cases) still passes; suffix-strip behavior verified against
hostile inputs (whitespace, casing, mid-string occurrences must NOT
match — only trailing pattern).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>