Deep slate diagnosis — 20260831

• Games: 12 · hits 8 · misses 4 · winner accuracy 0.6667
• Mean total-run bias on losses: -2.25 (negative = under-projected)
• Mean team-run error on losses: 8.25 · mean Brier on losses: 0.327

Loss primary drivers

• Starting-pitcher projection was wrong: 3
• Backed the wrong side with real conviction: 1

High-level causal issue register

[HIGH] An intermediate stage materially degraded winner direction

• Domain: pipeline_stages · affected: BAL@COL, NYY@LAA
• Evidence: historical_ml → variance_applied changed winner accuracy by -0.1667 (-2 hits)
• Action: List the exact games whose winner flipped at this stage, and keep both the before-stage and after-stage picks in the trace.
• Do not infer: Even if a later stage recovers the pick, this unstable step is still a risk; do not treat it as harmless.

[HIGH] Published margins are compressed relative to realized margins

• Domain: score_distribution · affected: DET@MIN, MIL@CHC, NYY@LAA, OAK@TEX, SD@CIN, SF@ATL
• Evidence: 11/12 predictions were ≤2 runs; 6/12 actual margins were ≥4; mean predicted margin 1.42 vs actual 4.92
• Action: Check whether predicted margins are too narrow across several recent out-of-sample slates (measuring tail coverage and margin MAE) before widening the score spread.
• Do not infer: A single blowout game is not a reason to widen every score prediction.

[HIGH] Published confidence buckets were poorly aligned on this slate

• Domain: calibration · affected: BAL@COL, CHW@HOU, DET@MIN, MIA@WAS, MIL@CHC, NYM@TB, NYY@LAA, OAK@TEX, PHI@ARI, SD@CIN, SEA@BOS, SF@ATL
• Evidence: weighted absolute calibration error 0.1661 across 12 games
• Action: Watch the same confidence buckets over the full rolling window before changing any calibration weight.
• Do not infer: A single slate has far too few games to establish a stable calibration curve.

[HIGH] Validated run models explain very little holdout variance

• Domain: model_validation · affected: BAL@COL, CHW@HOU, DET@MIN, MIA@WAS, MIL@CHC, NYM@TB, NYY@LAA, OAK@TEX, PHI@ARI, SD@CIN, SEA@BOS, SF@ATL
• Evidence: home/away/total R² = 0.0378/0.0275/0.0272
• Action: Treat exact run scores as low-confidence estimates, and look for stronger features or a better target before tuning the score tails.
• Do not infer: A small positive R² only means the model beats guessing the average; it does not make the exact scores reliable.

[INFO] Every settled game received structural QA

• Domain: coverage · affected: BAL@COL, CHW@HOU, DET@MIN, MIA@WAS, MIL@CHC, NYM@TB, NYY@LAA, OAK@TEX, PHI@ARI, SD@CIN, SEA@BOS, SF@ATL
• Evidence: 12/12 games structurally audited
• Action: Keep the per-game audit receipt in every report.
• Do not infer: Full audit coverage does not cancel the findings raised inside the audited games.

[INFO] Material calibrated-to-published movements have recorded lineage

• Domain: probability_lineage · affected: CHW@HOU, DET@MIN, MIA@WAS, MIL@CHC, NYY@LAA, OAK@TEX, SEA@BOS, SF@ATL
• Evidence: 8 material movement(s), all explained by named score-coherence and/or recency transformations
• Action: Keep showing every probability layer and the named transformation between them.
• Do not infer: A published probability above the calibrated one is not a governor bypass when a recorded transformation explains it.

[INFO] Simulation produced no winner-direction intervention on this slate

• Domain: simulation · affected: BAL@COL, CHW@HOU, DET@MIN, MIA@WAS, MIL@CHC, NYM@TB, NYY@LAA, OAK@TEX, PHI@ARI, SD@CIN, SEA@BOS, SF@ATL
• Evidence: flip candidates 0; published flips 0
• Action: Measure the simulation's contribution across rolling slates, separating score realization from winner flipping.
• Do not infer: Zero winner flips is not a defect and does not mean the simulation did no score-reconciliation work.

Cross-slate trends (persistent-pattern memory)

• Window: 20260828, 20260829, 20260830, 20260831 (3 prior slates loaded)
• Highest-confidence pick loss streak: 0 slate(s)
• Actionable pattern detected: True

Escalated persistent patterns

• simulation_inert_on_slate — 4 consecutive slates (4 in window)
• Target lever(s): simulation.allow_score
• Direction: confirm the simulation module is running, not silently no-oping
• Guardrail: simulation_inert_on_slate has persisted 4 consecutive slates; this is no longer single-slate noise. Treat it as an actionable pattern and raise the code/diagnostic action named; do not force a parameter move.
• weak_run_model_signal — 4 consecutive slates (4 in window)
• Target lever(s): training.xgb.n_estimators, training.xgb.max_depth, training.xgb.learning_rate
• Direction: investigate features/target via isolated training backtest (manual)
• Guardrail: weak_run_model_signal has persisted 4 consecutive slates; this is no longer single-slate noise. Treat it as an actionable pattern and raise the code/diagnostic action named; do not force a parameter move.

Metric direction across the window

• winner_accuracy: flat · series [0.4667, 0.9412, 0.5, 0.6667]
• weighted_calibration_error: improving · series [0.2179, 0.3961, 0.0693, 0.1661]
• margin_bias: regressing · series [-1.333, -1.882, -2.786, -3.5]
• mean_team_run_error_on_losses: regressing · series [5.0, 3.0, 7.0, 8.25]

Pipeline stage quality

• historical_ml: winner 8/12 (0.6667), team-run MAE 3.1283, margin MAE 4.7023
• variance_applied: winner 6/12 (0.5), team-run MAE 3.125, margin MAE 4.75
• enhanced_prediction: winner 8/12 (0.6667), team-run MAE 3.4583, margin MAE 4.75
• embedded_savant_run_environment: winner 8/12 (0.6667), team-run MAE 3.4167, margin MAE 4.6667
• fundamental_edge_target_review: winner 8/12 (0.6667), team-run MAE 3.4167, margin MAE 4.6667
• capped_target: winner 8/12 (0.6667), team-run MAE 3.4167, margin MAE 4.6667
• simulation_result: winner 8/12 (0.6667), team-run MAE 3.4167, margin MAE 4.6667
• published_final: winner 8/12 (0.6667), team-run MAE 3.4167, margin MAE 4.6667

Probability lineage

• BAL@COL: raw 0.7585 → calibrated 0.7435 → published 0.7492; score_coherence aligned probability to the published score margin
• CHW@HOU: raw 0.5085 → calibrated 0.5082 → published 0.53; score coherence followed by matchup-recency guardrail
• DET@MIN: raw 0.5085 → calibrated 0.5082 → published 0.53; score coherence followed by matchup-recency guardrail
• MIA@WAS: raw 0.5085 → calibrated 0.5082 → published 0.53; score coherence followed by matchup-recency guardrail
• MIL@CHC: raw 0.5085 → calibrated 0.5082 → published 0.53; score coherence followed by matchup-recency guardrail
• NYM@TB: raw 0.67 → calibrated 0.661 → published 0.6655; score_coherence aligned probability to the published score margin
• NYY@LAA: raw 0.5943 → calibrated 0.591 → published 0.5509; score coherence followed by matchup-recency guardrail
• OAK@TEX: raw 0.5085 → calibrated 0.5082 → published 0.53; score coherence followed by matchup-recency guardrail
• PHI@ARI: raw 0.5085 → calibrated 0.5082 → published 0.5269; score coherence followed by matchup-recency guardrail
• SD@CIN: raw 0.6947 → calibrated 0.677 → published 0.6735; score_coherence aligned probability to the published score margin
• SEA@BOS: raw 0.585 → calibrated 0.5816 → published 0.5602; score coherence followed by matchup-recency guardrail
• SF@ATL: raw 0.5085 → calibrated 0.5082 → published 0.53; score coherence followed by matchup-recency guardrail

Every game, in detail

BAL@COL — ✅ hit

• Predicted BAL 7-4 (conf -0.50, strong) · actual BAL 2-1
• Root cause: right_winner_wrong_score
• Errors: winner correct · Brier 0.06288 · team-run err 8.0 · total bias +8
• Contributing factors:
• total runs over-projected by 8 (11 vs 3)
• BAL offense over-projected by 5
• COL offense over-projected by 3

CHW@HOU — ✅ hit

• Predicted HOU 4-5 (conf +0.06, lean) · actual HOU 3-6
• Root cause: solid_correct_call
• Errors: winner correct · Brier 0.2209 · team-run err 2.0 · total bias +0

DET@MIN — ✅ hit

• Predicted MIN 4-5 (conf +0.06, lean) · actual MIN 1-11
• Root cause: right_winner_wrong_score
• Errors: winner correct · Brier 0.2209 · team-run err 9.0 · total bias -3
• Contributing factors:
• total runs under-projected by 3 (9 vs 12)
• DET offense over-projected by 3
• MIN offense under-projected by 6
• Taj Bradley (MIN) allowed 3 fewer ER than projected (1 vs 4)

MIA@WAS — ✅ hit

• Predicted WAS 3-4 (conf +0.06, lean) · actual WAS 3-6
• Root cause: solid_correct_call
• Errors: winner correct · Brier 0.2209 · team-run err 2.0 · total bias -2
• Contributing factors:
• Ryan Gusto (MIA) allowed 3 fewer ER than projected (1 vs 4)

MIL@CHC — ❌ miss

• Predicted MIL 6-5 (conf -0.06, lean) · actual CHC 3-17
• Root cause: Starting-pitcher projection was wrong
• Secondary cause: Run environment far higher than projected (under-dispersed)
• Errors: winner WRONG · Brier 0.2809 · team-run err 15.0 · total bias -9
• Contributing factors:
• total runs under-projected by 9 (11 vs 20)
• MIL offense over-projected by 3
• CHC offense under-projected by 12
• Kyle Harrison (MIL) allowed 6 more ER than projected (9 vs 3)

NYM@TB — ❌ miss

• Predicted TB 3-5 (conf +0.33, strong) · actual NYM 3-2
• Root cause: Starting-pitcher projection was wrong
• Secondary cause: Backed the wrong side with real conviction
• Errors: winner WRONG · Brier 0.44289 · team-run err 3.0 · total bias +3
• Contributing factors:
• total runs over-projected by 3 (8 vs 5)
• TB offense over-projected by 3
• Robert Stock (NYM) allowed 4 fewer ER than projected (0 vs 4)
• Ian Seymour (TB) allowed 3 more ER than projected (3 vs 0)
• committed to TB at 0.33 confidence and it lost

NYY@LAA — ❌ miss

• Predicted NYY 5-4 (conf -0.10, moderate) · actual LAA 1-10
• Root cause: Backed the wrong side with real conviction
• Secondary cause: Small edge on the losing side
• Errors: winner WRONG · Brier 0.30346 · team-run err 10.0 · total bias -2
• Contributing factors:
• NYY offense over-projected by 4
• LAA offense under-projected by 6
• committed to NYY at 0.10 confidence and it lost

OAK@TEX — ✅ hit

• Predicted TEX 3-5 (conf +0.06, lean) · actual TEX 1-8
• Root cause: right_winner_wrong_score
• Errors: winner correct · Brier 0.2209 · team-run err 5.0 · total bias -1
• Contributing factors:
• TEX offense under-projected by 3
• Gage Jump (OAK) allowed 5 more ER than projected (7 vs 2)

PHI@ARI — ✅ hit

• Predicted PHI 5-4 (conf -0.05, lean) · actual PHI 2-1
• Root cause: right_winner_wrong_score
• Errors: winner correct · Brier 0.223783 · team-run err 6.0 · total bias +6
• Contributing factors:
• total runs over-projected by 6 (9 vs 3)
• PHI offense over-projected by 3
• ARI offense over-projected by 3
• Aaron Nola (PHI) allowed 3 fewer ER than projected (1 vs 4)

SD@CIN — ✅ hit

• Predicted SD 6-4 (conf -0.35, strong) · actual SD 5-0
• Root cause: right_winner_wrong_score
• Errors: winner correct · Brier 0.1066 · team-run err 5.0 · total bias +5
• Contributing factors:
• total runs over-projected by 5 (10 vs 5)
• CIN offense over-projected by 4
• Michael King (SD) allowed 3 fewer ER than projected (0 vs 3)

SEA@BOS — ✅ hit

• Predicted BOS 2-3 (conf +0.12, moderate) · actual BOS 8-9
• Root cause: right_winner_wrong_score
• Errors: winner correct · Brier 0.193436 · team-run err 12.0 · total bias -12
• Contributing factors:
• total runs under-projected by 12 (5 vs 17)
• SEA offense under-projected by 6
• BOS offense under-projected by 6
• George Kirby (SEA) allowed 3 more ER than projected (5 vs 2)
• Payton Tolle (BOS) allowed 4 more ER than projected (4 vs 0)

SF@ATL — ❌ miss

• Predicted ATL 4-5 (conf +0.06, lean) · actual SF 7-3
• Root cause: Starting-pitcher projection was wrong
• Secondary cause: Small edge on the losing side
• Errors: winner WRONG · Brier 0.2809 · team-run err 5.0 · total bias -1
• Contributing factors:
• SF offense under-projected by 3
• Bryce Elder (ATL) allowed 4 more ER than projected (5 vs 1)

Cross-game lever tension

Each candidate fix, with the games it would help and the games it would hurt. A lever that repairs one game and breaks two is visible here before it is proposed.

score.variance_multiplier / score.variance_floor (widen run distributions)

• Targets: brier_score, calibration_error
• Net: 2 helped vs 2 hurt
• Would help:
• NYM@TB (miss): lost a 0.33-confidence pick; wider tails lower overconfidence
• NYY@LAA (miss): lost a 0.10-confidence pick; wider tails lower overconfidence
• Would hurt / risk:
• BAL@COL (hit): strong correct call at 0.50; widening softens a good edge
• SD@CIN (hit): strong correct call at 0.35; widening softens a good edge

player.pitcher.v4 form weights / player.pitcher_prior_games

• Targets: starter_er_mae, starter_outs_mae
• Net: 9 helped vs 1 hurt
• Would help:
• DET@MIN (hit): starter ER off by 3
• MIA@WAS (hit): starter ER off by 3
• MIL@CHC (miss): starter ER off by 6
• NYM@TB (miss): starter ER off by 4
• OAK@TEX (hit): starter ER off by 5
• PHI@ARI (hit): starter ER off by 3
• SD@CIN (hit): starter ER off by 3
• SEA@BOS (hit): starter ER off by 4
• SF@ATL (miss): starter ER off by 4
• Would hurt / risk:
• BAL@COL (hit): starter projection was accurate (ER off 1); reweight risks disturbing it

recent. / season. run-mean weighting

• Targets: team_run_mae, total_run_mae, accuracy
• Net: 9 helped vs 2 hurt
• Would help:
• BAL@COL (hit): team runs off by 8
• DET@MIN (hit): team runs off by 9
• MIL@CHC (miss): team runs off by 15
• NYY@LAA (miss): team runs off by 10
• OAK@TEX (hit): team runs off by 5
• PHI@ARI (hit): team runs off by 6
• SD@CIN (hit): team runs off by 5
• SEA@BOS (hit): team runs off by 12
• SF@ATL (miss): team runs off by 5
• Would hurt / risk:
• CHW@HOU (hit): team runs already close (off 2); reweight risks disturbing it
• MIA@WAS (hit): team runs already close (off 2); reweight risks disturbing it