Architecture
01 · Core architecture
PI, reader.py driving, sources.py deterministic fetch, engine.py validation, four engine routes, Ledger and PI_REVIEW.
Open chart →Workflow
02 · Reading loop
feed fetching, predict-first guessing, read three sentences + verbatim quotes, attach, grow into rows, decide, PI verdict.
Open chart →Lifecycle
03 · Disagreement lifecycle
Paper → candidate → disagreement row → XMM-decidable → PI card → exported proposal lane, plus stale/semantic terminal states.
Open chart →Data Flow
04 · Ledger data flow
ADS/arXiv fetching → reading_text → read → verbatim validation → attach → papers/rows/queue persistence.
Open chart →Sequence
05 · Engine routing and quota
ai-quota gating → nim/luna/sonnet reading seats → opus structural adjudication → engine.call validation.
Open chart →Architecture
06 · Honesty rules
Verbatim quotes, bibcodes from ADS, two papers per row, four open classes, prediction thermometer, inventory ≠ disagreement.
Open chart →