ChatLedger
Personal conversation analysis system. Extracts structured data (claims, decisions, emotional arcs, negotiation patterns) from SMS and chat history, with a benchmarked extraction schema refined through multi-metric field screening.
- Benchmarked V3 extraction schema with weighted field scoring
- Four-metric candidate screening (consistency, information gain, discrimination, redundancy)
- Calibration measurement for extraction confidence
Activity Timeline
- Adversarial review launched for vtuber-radio feature; tailnet security validated.
Sol dispatched for M1–M3 xhigh passes alongside adversarial read of 6 core files. Tailnet grant security empirically confirmed before the review concluded.
- Benchmark expansion pilot complete; GPT-5.6 Sol ran all 5 chunks successfully.
Pilot phase closed. Mission framework established. Subsequent runs timed out or pending; status report written.
- ~20 stalled projects diagnosed: cache-cold handoff failures at decision points.
Projects stalling at 70-80% completion traced to sessions losing context when cache expires at key decision points. Automated summary and handoff mechanism scoped as the fix; no implementation started.
- Background benchmarks done; leaked auth token confirmed revoked; moved to paper phase.
P4G benchmark across four models and haiku45 enrichment with Gemini retry both completed. Auth token from .codex/auth.json verified revoked with no residual exposure. Project transitions from verification into paper creation.