v0.23.0 — ### Added

• Poll-now button on each watchlist row — calls POST /api/watched-reports/{code}/poll which wraps ingest_report + ingest_events_for_report with proper commits and per-call error capture, so users never hit the half-committed-session footgun (the bug that broke FRU dev ingest twice during T-103/T-104). Synchronous: returns only after the ingest finishes (or fails). UI shows a per-row status snippet ("ingested · fights+208 · events+29725" / "already complete · no work" / error message).
• jobs/poll_watched.py refactored: _poll_one_row() extracted as the canonical ingest-one-report path; poll_once() and the new poll_one_by_code() both go through it. 50-line dedup.
• 2 new API tests for the Poll-now endpoint (404 when not watching; skips-complete behavior). 192 tests total.

Changed

• DEFAULT_ENCOUNTERS in jobs/backfill_field.py expanded to include TOP (1068) and DSR (1065) alongside FRU (1079) + current Savage tier (101–105). Verified encounter IDs against FFLogs' worldData.encounter(id).name. Other ultimates (TEA=1075, UWU=1074, UCoB=1073) documented in the comment for one-line future addition.
• FieldStats.jsx encounter-name lookup table updated to render TOP / DSR instead of raw IDs.

Scheduled (no code)

• T-109 — Tighten T-004's combatant filter for Ultimate reports. PLAN §11 Phase 1 table updated. Captured in IDEAS.md: T-004's type == "Player" filter on masterData.actors lets ~12k NPC actors leak through per Ultimate report (verified on FRU: 208 fights × ~12k = 2.5M combatant rows). Fix: intersect masterData with the per-fight active-source set during ingest, dropping rows that never appear as source_id in cast/damage events. Workarounds in T-203/T-206 still function after the fix; they just become redundant.