2026-05-24

v0.20.0 — ### Added

- **T-202** Cross-group consensus → `fight_model` persistence. `analysis/consensus.py::write_consensus_to_fight_model(session, encounter_id, *, version=1)` runs T-104, then replac…

T-202 Cross-group consensus → fight_model persistence. analysis/consensus.py::write_consensus_to_fight_model(session, encounter_id, *, version=1) runs T-104, then replaces rows for (encounter_id, version) atomically. seq is assigned by sorting canonical abilities by median_relative_t_ms within each phase. confidence = occurrence_rate; type_label left null (T-203 will fill: raidwide/tankbuster/tower/spread/stack/tether/enrage). meta carries {sample_count, pulls_reaching} for downstream filtering.
read_fight_model(session, encounter_id) mirrors the write shape for the API.
API: POST /api/encounters/{id}/fight-model/persist (trigger a write), GET /api/encounters/{id}/fight-model (read persisted rows).
5 new tests covering write/read/idempotency/seq-ordering/empty cases. 176 tests total.
Live AC: persisted FRU (encounter 1079) — 6 phases / 149 canonical abilities / 11 pulls aggregated — and M5S (encounter 101) — 2 phases / 18 abilities / 19 pulls.