Genealogy Research
AI-assisted family history research using multiple search APIs, document analysis, and systematic brick-wall breaking strategies.
- Multi-source search (Exa, Brave, Codex)
- Document analysis with Gemini OCR
- Brick-wall breaker agent for stuck research
- Family tree builder with relationship integrity
Activity Timeline
- Report download deployed (.md + .pdf, auth-gated, 43/43 tests passing).
Headless Chrome export at 0.19–2.5s behind owner-credential gate. Covers byte identity, PDF structure, cache, concurrency, and auth cases.
- Sol review confirmed two P0 security findings; remediation queued.
Intent provenance relay-path review surfaced missing actor validation in jsonl-append.py and unauthenticated label handling in interview-app/server.py. Both verified P0; no fixes applied yet.
- Privacy audit: 5,144+ files swept, 2 edge-case collisions remediated.
Five sequential gate evaluation rounds completed with Sol-assisted adjudication in round 1. Privacy inventory updated with new coverage metrics.
- Privacy filter expanded 81→146 entries; 5,144-file sweep passed clean.
Diacritic folding and full-name pairs added to deny-list. Zero private-name leaks confirmed across 5,144 files. Five claims pending primary-source retrieval for adjudication.
- LAC-1938 adjudication: 3 gate failures, mid-loop on corrections.
Each round rejected for distinct data-quality issues: uncited claims on recheck, a date correction (Oct 29 → Jul 9 1938), and a claim scored without an underlying LAC record. Phase 4 pending Gazette browser lookup result.
- 3-round data quality audit; one LAC-index score frozen pending primary-source retrieval.
Independent Sol audit found 19 uncited claims in Round 1 (vs. 4 reported), plus arithmetic errors and missing metadata. Rounds 2–3 narrowed scope. Subject clm-033 LAC score frozen at E4 fallback (0.50) pending Round 4 browser retrieval.
- 3 adjudication rounds all failed validation; clm-033 frozen on missing source.
Round 1 found 19 uncited pairs (4 reported); full join re-run and sweep regeneration required. Rounds 2 and 3 applied corrections and froze respectively on a primary Gazette record requirement.
- 3 validation rounds; 106-profile sweep; clm-033 blocked on primary source.
Stale-assertion sweep promoted 8 claims and corrected profile count. the LAC-1938 identity adjudication blocked pending primary source retrieval for clm-033 evidentiary standard.
- Concurrency fix landed; 5 profiles advanced via documentary chains.
46/46 tests pass after serializing 43 Chrome processes via counting semaphore with SIGKILL recovery and FIFO ordering. Five profiles gained exact dates from register and obituary sources. Candidates below rubric threshold held pending; premature merge blocked.
- Browser automation concurrency fix delivered; 46/46 tests green; methodology pivoted to public-record system.
43-process Chrome contention hazard resolved with counting semaphore and critical section. Project reframed from personal-record correction to generalizable public-record validation with dual-search strategy (GPT Pro + Fable-with-Exa).
- Methodology upgraded to validation-based; book rc6→rc7 marked READY with TTS.
Moved from personal-knowledge assignment to public-record evidence chains (children/marriage/obituary). Book grew 30→33 pages with corrected TOC; 5:29 Kokoro narration generated and marked READY.
- Research loop rebuild initiated with tiered AI judgment engine.
GPT Pro for image transcription, codex-council for hard adjudications. 8+ record images uploaded and queued at resolutions up to 4363×3468 px.
- 8+ record images submitted for GPT Pro transcription; codex-council adjudication loop configured.
Transcription loop set up with scale factors per image. Codex-council adjudication planned for disputed readings. Session transcript truncated before outputs were confirmed; results pending.