Nightly self-improvement · slate 2026-09-27
Generated 2026-09-28T05:00:00.007835-04:00 ET from official finals (mlb-api). Every learning stage is listed with what it did. "Blocked" means it wanted to act and a safety rule stopped it; "nothing to learn" means it had no eligible evidence.
Slate
2026-09-27
status completed
Winners
9/14
Brier 0.2398
Learned or changed
0
of 5 stages
Blocked
1
wanted to act and could not
Release
v4.31.0
DRIFT
What each learning stage did
| Stage | Outcome | Detail |
|---|---|---|
| KPI settlement | failed | did not run |
| Bullpen usage model | nothing to learn | No pregame bullpen forecasts to settle. Refit: not attempted. |
| Adaptive parameter proposal (Luna) | blocked | quarantined_invalid_artifact: slate quarantined from new parameter proposals until critical/error artifact findings are corrected |
| Auto-apply | nothing to learn | no_promotable_candidate |
| Code agent | nothing to learn | disabled_by_config: code changes are made on the development host, never in the container. |
Games
| Game | Predicted | Actual | Home win prob | Result | Main driver | Starters |
|---|---|---|---|---|---|---|
| ARI@SD | 6-4 | 4-9 | 0.489 | miss | Genuine toss-up that flipped | Michael Soroka: 16 proj / 14 real outs; Randy Vásquez: 15 proj / 18 real outs |
| ATL@MIA | 4-5 | 3-5 | 0.505 | hit | JR Ritchie: 14 proj / 17 real outs; Janson Junk: 14 proj / 14 real outs | |
| CHC@BOS | 4-3 | 6-2 | 0.495 | hit | Shota Imanaga: 17 proj / 3 real outs; Tanner Houck: 3 proj / 9 real outs | |
| CIN@TOR | 4-5 | 1-5 | 0.512 | hit | Brandon Williamson: 13 proj / 9 real outs; Max Scherzer: 15 proj / 21 real outs | |
| CLE@KC | 6-4 | 2-3 | 0.440 | miss | Run environment far lower than projected (over-dispersed) | Parker Messick: 19 proj / 9 real outs; Daniel Lynch IV: 16 proj / 16 real outs |
| COL@CHW | 5-8 | 2-4 | 0.560 | hit | Kyle Freeland: 13 proj / 12 real outs; Anthony Kay: 14 proj / 15 real outs | |
| HOU@OAK | 8-6 | 9-0 | 0.440 | hit | Peter Lambert: 16 proj / 6 real outs; Seth Johnson: 12 proj / 5 real outs | |
| LAA@SEA | 3-5 | 3-7 | 0.560 | hit | Yusei Kikuchi: 15 proj / 15 real outs; Logan Gilbert: 17 proj / 12 real outs | |
| LAD@SF | 6-3 | 5-1 | 0.420 | hit | Jack Dreyer: 15 proj / 6 real outs; Carson Seymour: 12 proj / 3 real outs | |
| NYM@WAS | 7-5 | 4-6 | 0.469 | miss | Small edge on the losing side | Sean Manaea: 17 proj / 18 real outs; DJ Herz: 11 proj / 9 real outs |
| PIT@DET | 4-3 | 4-2 | 0.495 | hit | Jared Jones: 17 proj / 12 real outs; River Ryan: 9 proj / 4 real outs | |
| STL@MIL | 2-5 | 4-6 | 0.580 | hit | Andre Pallante: 16 proj / 15 real outs; Jacob Misiorowski: 18 proj / 16 real outs | |
| TB@PHI | 4-3 | 3-7 | 0.424 | miss | Backed the wrong side with real conviction | Nick Martinez: 18 proj / 8 real outs; Zack Wheeler: 18 proj / 18 real outs |
| TEX@MIN | 5-4 | 4-6 | 0.488 | miss | Genuine toss-up that flipped | MacKenzie Gore: 14 proj / 7 real outs; Dean Kremer: 15 proj / 21 real outs |
Day over day
Rolling evidence window against the prior night. Associations, not causal proof.
| Metric | Prior | Current | Change | Result |
|---|---|---|---|---|
| brier_score | 0.2413 | 0.2462 | 0.0050 | regressed |
| accuracy | 0.5932 | 0.5726 | -0.0206 | regressed |
| team_run_mae | 2.6314 | 2.5684 | -0.0630 | improved |
| total_run_mae | 3.8559 | 3.8034 | -0.0525 | improved |
| calibration_error | 0.0876 | 0.0862 | -0.0014 | improved |
| starter_er_mae | 1.7895 | 1.7965 | 0.0070 | regressed |
| starter_outs_mae | 3.9123 | 4.0442 | 0.1319 | regressed |
Calibration produced for the next slate
| Metric | Scale used | Measured | Rows | Status |
|---|---|---|---|---|
| No calibration produced. | ||||
Luna KPI candidates and their promotion gate
Promotion needs 400+ prospective rows over 5+ slates after the proposal, with a gain interval above zero. Only verified pregame forecasts count; local replays never do.
| KPI | Status | Target | Proposed | Rows | Slates | Gain interval | Progress |
|---|---|---|---|---|---|---|---|
| luna_pitcher_release_speed_vs_righthanded_starters | shadow | SO | 20260923 | 330/400 | 5/5 | [-0.0007, 0.0007] | |
| luna_pitcher_barrel_rate_vs_rhh_righthanded_starters | shadow | HR | 20260923 | 330/400 | 5/5 | [-0.0004, 0.0049] | |
| luna_batter_contact_rate_vs_breaking_righthanded_starters | shadow | H | 20260923 | 330/400 | 5/5 | [-0.0045, 0.0044] | |
| luna_batter_contact_rate_vs_breaking_left_handed_starters_for_strikeouts | shadow | SO | 20260924 | 122/400 | 4/5 | [-0.0027, 0.0021] | |
| luna_pitcher_contact_xwoba_allowed_vs_righthanded_starters_for_strikeouts | shadow | SO | 20260924 | 122/400 | 4/5 | [-0.0001, 0.0000] | |
| luna_pitcher_whiff_rate_breaking_vs_left_handed_hitters_for_hits | shadow | H | 20260924 | 122/400 | 4/5 | [-0.0056, -0.0011] | |
| luna_pitcher_home_run_rate_vs_righthanded_starters | shadow | HR | 20260925 | 122/400 | 4/5 | [-0.0010, 0.0056] | |
| luna_batter_ground_ball_rate_vs_fastball_vs_lhp_for_home_runs | shadow | HR | 20260926 | 86/400 | 3/5 | [-0.0001, -0.0000] | |
| luna_pitcher_whiff_rate_vs_right_handed_starters_for_batter_strikeouts | shadow | SO | 20260929 | 51/400 | 2/5 | [–, –] | |
| luna_batter_whiff_rate_vs_breaking_right_handed_starters_for_hits | shadow | H | 20260929 | 51/400 | 2/5 | [–, –] | |
| luna_batter_walk_rate_vs_righthanded_pitchers_for_home_runs | retired | HR | 20260924 | 0/400 | 0/5 | [–, –] |
Operator actions the cycle asked for (0)
Changes to code or a model recipe are never made by the nightly cycle. These are the measured findings that need one.
| Priority | Owner | Finding | Action |
|---|---|---|---|
| No operator actions requested. | |||
Luna post-mortem
Luna (gpt-5.6-luna) post-mortem, status received. Full analysis.
# 1. Evidence receipt and gaps ## Sources received All seven `source_manifest` entries were checked: | Source | Status | Purpose | |---|---:|---| | `prediction_review_20260927.json` | Loaded | Official outcomes and scoring errors | | `PREDICTION_SYSTEM_DEEP_AUDIT_20260927.json` | Loaded | Probability, provenance, starter, and decision contracts | | `continuous_learning_daily_20260927.json` | Loaded | Stage quality, calibration, probability lineage, score distribution, and starter comparisons | | `detailed_20260927.html` | Loaded | Full game and simulation rendering | | `20260927_ultra_detailed_player_audit.html` | Loaded | Player-level prediction trace | | `audit_report_20260927.html` | Loaded | Invariant findings | | `audit_report_20260927_integrity_revision.html` | **Missing** | Post-Luna remediation proof for immutable revision artifacts | The missing integrity-revision audit is material to remediation verification, although it is not evidence that the prediction model itself failed. ## Games reviewed The `games` array contains 14 games, all reviewed: ARI@SD, ATL@MIA, CHC@BOS, CIN@TOR, CLE@KC, COL@CHW, HOU@OAK, LAA@SEA, LAD@SF, NYM@WAS, PIT@DET, STL@MIL, TB@PHI, and TEX@MIN. The game-level results reconcile to 9 winner hits and 5 misses. However, the evidence contains multiple coverage contradictions: - `games_reviewed: 14`, `hits: 9`, and `continuous_learning_evidence.summary` reports 14 settled games. - `coverage_matrix.expected_games: 14`. - But `canonical_arti