update_game_started (started_at, num_players, num_rounds, player_ids)
was defined in event_store but had zero callers. 289/289 staging games
had those fields NULL — queries that joined on them returned garbage,
and the denormalized player_ids GIN index was dead weight.
log_game_start now calls create_game THEN update_game_started in one
async task. If create fails, update is skipped (row doesn't exist).
handlers.py passes num_rounds and player_ids through at call time.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When handle_player_leave emptied a room or handle_end_game was invoked,
the room was removed from memory without touching games_v2. Periodic
cleanup only scans in-memory rooms, so those rows were stranded as
status='active' forever — staging had 42 orphans accumulated over 5h.
- event_store.update_game_abandoned: guarded UPDATE (status='active' only)
- GameLogger.log_game_abandoned{,_async}: fire-and-forget wrapper
- handle_end_game + handle_player_leave: flip status before remove_room
- LEADERBOARD_INCLUDE_TEST_DEFAULT: env override so staging can show
soak-harness accounts by default; prod keeps them hidden
Verified on staging: 42 orphans swept on restart, soak accounts now
visible on /api/stats/leaderboard (rank 1-4).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
