Deep slate diagnosis — 20260921

• Games: 3 · hits 1 · misses 2 · winner accuracy 0.3333
• Mean total-run bias on losses: 2.0 (negative = under-projected)
• Mean team-run error on losses: 3.0 · mean Brier on losses: 0.289

Loss primary drivers

• Starting-pitcher projection was wrong: 1
• Genuine toss-up that flipped: 1

High-level causal issue register

[CRITICAL] Published probability differs materially without a recorded transformation

• Domain: probability_lineage · affected: WAS@DET
• Evidence: 1 game(s) have an unexplained ≥2pp calibrated-to-published delta
• Action: Block any promotion until the missing probability transformation is recorded and reconciled.
• Do not infer: Do not assume the probability governor was bypassed until you have traced the pick-side math.

[HIGH] Blocking canonical artifact error

• Domain: artifact_integrity · affected: 20260921_WAS@DET
• Evidence: structural audit totals: {'critical': 0, 'error': 2, 'warning': 0, 'info': 22}
• Action: Exclude this slate from parameter learning until the artifact-generation step that wrote the broken file is fixed.
• Do not infer: A malformed artifact is a file bug, not proof the prediction model was wrong; do not blame accuracy changes on it.

[HIGH] An intermediate stage materially degraded winner direction

• Domain: pipeline_stages · affected: WAS@DET
• Evidence: historical_ml → variance_applied changed winner accuracy by -0.3333 (-1 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 confidence buckets were poorly aligned on this slate

• Domain: calibration · affected: MIN@SF, TOR@BAL, WAS@DET
• Evidence: weighted absolute calibration error 0.2028 across 3 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: MIN@SF, TOR@BAL, WAS@DET
• Evidence: home/away/total R² = 0.0187/0.0317/0.0089
• 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: MIN@SF, TOR@BAL, WAS@DET
• Evidence: 3/3 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] Simulation produced no winner-direction intervention on this slate

• Domain: simulation · affected: MIN@SF, TOR@BAL, WAS@DET
• 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: 20260911, 20260912, 20260913, 20260914, 20260915, 20260916, 20260917, 20260918, 20260919, 20260920, 20260921 (10 prior slates loaded)
• Highest-confidence pick loss streak: 1 slate(s)
• Actionable pattern detected: True

Escalated persistent patterns

• pipeline_winner_instability — 4 consecutive slates (5 in window)
• Target lever(s): code/diagnostic — no parameter
• Direction: preserve base_ml_winner before integer rounding (code fix)
• Guardrail: pipeline_winner_instability 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.
• simulation_inert_on_slate — 11 consecutive slates (11 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 11 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.
• single_slate_calibration_instability — 3 consecutive slates (8 in window)
• Target lever(s): confidence.empirical_max_weight, confidence.base_cap, confidence.moderate_cap
• Direction: increase empirical shrinkage
• Guardrail: single_slate_calibration_instability has persisted 3 consecutive slates; this is no longer single-slate noise. Treat it as an actionable pattern and propose a bounded shadow experiment against a target lever.
• weak_run_model_signal — 11 consecutive slates (11 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 11 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: regressing · series [0.6667, 0.5333, 0.6, 0.7, 0.4, 0.6667, 0.4444, 0.6667, 0.7333, 0.6, 0.3333]
• weighted_calibration_error: flat · series [0.2091, 0.1711, 0.0827, 0.2812, 0.2795, 0.1661, 0.1016, 0.1247, 0.3458, 0.3184, 0.2028]
• margin_bias: regressing · series [-2.933, -2.667, -1.8, -0.4, -1.667, -1.133, -2.111, -2.0, -0.667, -1.533, -2.667]
• mean_team_run_error_on_losses: improving · series [5.0, 6.143, 5.0, 6.333, 5.556, 6.4, 3.6, 5.2, 6.5, 6.5, 3.0]

Pipeline stage quality

• historical_ml: winner 1/3 (0.3333), team-run MAE 1.9153, margin MAE 3.7913
• variance_applied: winner 0/3 (0.0), team-run MAE 2.0, margin MAE 4.0
• enhanced_prediction: winner 1/3 (0.3333), team-run MAE 2.0, margin MAE 4.0
• embedded_savant_run_environment: winner 1/3 (0.3333), team-run MAE 2.0, margin MAE 4.0
• fundamental_edge_target_review: winner 1/3 (0.3333), team-run MAE 2.0, margin MAE 4.0
• capped_target: winner 1/3 (0.3333), team-run MAE 2.0, margin MAE 4.0
• simulation_result: winner 1/3 (0.3333), team-run MAE 2.0, margin MAE 4.0
• published_final: winner 1/3 (0.3333), team-run MAE 2.0, margin MAE 4.0

Probability lineage

• MIN@SF: raw 0.5702 → calibrated 0.5848 → published 0.5675; no recorded transformation explains the publication delta
• TOR@BAL: raw 0.5 → calibrated 0.5073 → published 0.5053; published probability matches the calibrated/governed value
• WAS@DET: raw 0.572 → calibrated 0.5774 → published 0.5358; no recorded transformation explains the publication delta

Every game, in detail

MIN@SF — ❌ miss

• Predicted MIN 5-4 (conf -0.14, moderate) · actual SF 2-5
• Root cause: Starting-pitcher projection was wrong
• Secondary cause: Backed the wrong side with real conviction
• Errors: winner WRONG · Brier 0.32206 · team-run err 4.0 · total bias +2
• Contributing factors:
• MIN offense over-projected by 3
• Zebby Matthews (MIN) allowed 4 more ER than projected (5 vs 1)
• committed to MIN at 0.14 confidence and it lost

TOR@BAL — ❌ miss

• Predicted TOR 5-4 (conf -0.01, toss_up) · actual BAL 3-4
• Root cause: Genuine toss-up that flipped
• Errors: winner WRONG · Brier 0.255285 · team-run err 2.0 · total bias +2

WAS@DET — ✅ hit

• Predicted DET 4-5 (conf +0.07, lean) · actual DET 2-9
• Root cause: right_winner_wrong_score
• Errors: winner correct · Brier 0.215473 · team-run err 6.0 · total bias -2
• Contributing factors:
• DET offense under-projected by 4
• DJ Herz (WAS) allowed 3 more ER than projected (4 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: 1 helped vs 0 hurt
• Would help:
• MIN@SF (miss): lost a 0.14-confidence pick; wider tails lower overconfidence

player.pitcher.v4 form weights / player.pitcher_prior_games

• Targets: starter_er_mae, starter_outs_mae
• Net: 2 helped vs 0 hurt
• Would help:
• MIN@SF (miss): starter ER off by 4
• WAS@DET (hit): starter ER off by 3

recent. / season. run-mean weighting

• Targets: team_run_mae, total_run_mae, accuracy
• Net: 2 helped vs 1 hurt
• Would help:
• MIN@SF (miss): team runs off by 4
• WAS@DET (hit): team runs off by 6
• Would hurt / risk:
• TOR@BAL (miss): team runs already close (off 2); reweight risks disturbing it