- fn.COUNT(fn.DISTINCT(expr)) → fn.COUNT(expr.distinct()) for correct
COUNT(DISTINCT ...) SQL on PostgreSQL
- _get_player_pairs() now also scans Decision table to include pitchers
who have a Decision row but no StratPlay rows (rare edge case)
- Updated stale docstring references to PlayerSeasonStats and r.k
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The previous approach accumulated per-game deltas into season stats rows,
which was fragile — partial processing corrupted stats, upsert bugs
compounded, and there was no self-healing mechanism.
Now update_season_stats() recomputes full season totals from all StratPlay
rows for each affected player whenever a game is processed. The result
replaces whatever was stored, eliminating double-counting and enabling
self-healing via force=True.
Also fixes:
- evolution_evaluator.py: broken PlayerSeasonStats import → queries
BattingSeasonStats or PitchingSeasonStats based on card_type
- evolution_evaluator.py: r.k → r.strikeouts
- test_evolution_models.py, test_postgame_evolution.py: PlayerSeasonStats
→ BattingSeasonStats (model never existed)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Mirror the batter_id is None guard in _build_pitching_groups() so that
a StratPlay row with a null pitcher_id is skipped rather than creating
a None key in the groups dict (which would fail on the NOT NULL FK
constraint during upsert).
- Revert Dockerfile to the next-release base: drop the COPY path change
and CMD addition that were already merged in PR #101 and are unrelated
to the ProcessedGame ledger feature.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Closes#105
Replace the last_game FK guard in update_season_stats() with an atomic
INSERT into a new processed_game ledger table. The old guard only blocked
same-game immediate replay; it was silently bypassed if game G+1 was
processed first (last_game already overwritten). The ledger is keyed on
game_id so any re-delivery — including out-of-order — is caught reliably.
Changes:
- app/db_engine.py: add ProcessedGame model (game FK PK + processed_at)
- app/services/season_stats.py: replace last_game check with
ProcessedGame.get_or_create(); import ProcessedGame; update docstrings
- migrations/2026-03-18_add_processed_game.sql: CREATE TABLE IF NOT EXISTS
processed_game with FK to stratgame ON DELETE CASCADE
- tests/conftest.py: add ProcessedGame to imports and _TEST_MODELS list
- tests/test_season_stats_update.py: add test_out_of_order_replay_prevented;
update test_double_count_prevention docstring
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Correct idempotency guard docstring in update_season_stats() to
accurately describe the last_game FK check limitation: only detects
replay of the most-recently-processed game; out-of-order re-delivery
(game G after G+1) bypasses the guard. References issue #105 for the
planned ProcessedGame ledger fix.
- Fix migration card_type comment: 'batting' or 'pitching' → 'batter',
'sp', or 'rp' to match actual seeded values.
- Remove local rarity fixture in test_season_stats_update.py that
shadowed the conftest.py fixture; remove unused rarity parameter from
player_batter and player_pitcher fixtures.
- Update test_double_count_prevention docstring to note the known
out-of-order re-delivery limitation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- CRITICAL: Fix migration FK refs player(id) → player(player_id)
- Remove dead is_start flag from pitching groups (no starts column)
- Fix hr → homerun in test make_play helper
- Add explanatory comment to ruff.toml
- Replace print() with logging in seed script
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Implement update_season_stats(game_id) in app/services/season_stats.py.
Aggregates StratPlay batting/pitching stats and Decision win/loss/save
data into PlayerSeasonStats with idempotency guard and dual-backend
upsert (PostgreSQL EXCLUDED increments, SQLite read-modify-write).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
