Deep slate report 20260914

Which reports actually fed the diagnosis. A source shown as missing was not silently assumed.

High-level causal issue register

Source	Status	Evidence supplied	Path	Bytes
prediction review	loaded	official outcomes and scoring errors	/app/output/reports/prediction_review_20260914.json	256117
forensic deep audit	loaded	probability, provenance, starter and decision contracts	/app/output/reports/PREDICTION_SYSTEM_DEEP_AUDIT_20260914.json	166036
continuous learning	loaded	pipeline-stage quality, calibration, probability lineage, score distribution, and official starter comparisons	/app/output/learning/continuous_learning_daily_20260914.json	1715357
detailed report	loaded	full game and simulation rendering	/app/output/reports/detailed_20260914.html	240489
ultra detailed player audit	loaded	player-level prediction trace	/app/output/reports/20260914_ultra_detailed_player_audit.html	22980257
structural audit	loaded	all invariant findings	/app/output/reports/audit_report_20260914.html	1234184
integrity revision audit	missing	post-Luna remediation proof for immutable revision artifacts	/app/output/reports/audit_report_20260914_integrity_revision.html	n/a

Deterministic findings across data integrity, model validation, calibration, score distribution, pipeline stages, simulation, and coverage. Critical is reserved for broken evidence or an unexplained contract violation; high marks material model/process risk. Each row pairs the action to take (Do this) with the wrong conclusion to avoid (Don't conclude).

Cross-slate trends (persistent-pattern memory)

Escalated persistent patterns

A finding that has repeated at or beyond the escalation streak is no longer single-slate noise. It carries its target registered lever and an action-oriented guardrail.

Finding streaks across the window

Metric direction across the window

Pipeline stage regression

Every stored stage is scored against the same official finals. A later recovery remains visible; it does not erase an unstable intermediate flip.

Probability lineage and calibration

All columns are pick-side probabilities, so away favorites are not accidentally compared using home-side semantics.

Score distribution and margin compression

Narrow predicted margins next to wide actual margins is the signature the score-variance tail is too tight.

Per-game evidence coverage

Every game, in depth (10)

Severity	Domain	Finding and evidence	Affected	Action & guardrail
CRITICAL	probability_lineage	Published probability differs materially without a recorded transformation 2 game(s) have an unexplained ≥2pp calibrated-to-published delta	BAL@NYM, MIA@ARI	Do thisBlock any promotion until the missing probability transformation is recorded and reconciled. Don't concludeDo not assume the probability governor was bypassed until you have traced the pick-side math.
HIGH	calibration	Published confidence buckets were poorly aligned on this slate weighted absolute calibration error 0.2812 across 10 games	ATL@CHC, BAL@NYM, CHW@CLE, DET@TOR, LAD@CIN, MIA@ARI, NYY@MIN, SD@COL, SEA@LAA, SF@STL	Do thisWatch the same confidence buckets over the full rolling window before changing any calibration weight. Don't concludeA single slate has far too few games to establish a stable calibration curve.
HIGH	model_validation	Validated run models explain very little holdout variance home/away/total R² = 0.0118/0.0210/0.0207	ATL@CHC, BAL@NYM, CHW@CLE, DET@TOR, LAD@CIN, MIA@ARI, NYY@MIN, SD@COL, SEA@LAA, SF@STL	Do thisTreat exact run scores as low-confidence estimates, and look for stronger features or a better target before tuning the score tails. Don't concludeA small positive R² only means the model beats guessing the average; it does not make the exact scores reliable.
INFO	coverage	Every settled game received structural QA 10/10 games structurally audited	ATL@CHC, BAL@NYM, CHW@CLE, DET@TOR, LAD@CIN, MIA@ARI, NYY@MIN, SD@COL, SEA@LAA, SF@STL	Do thisKeep the per-game audit receipt in every report. Don't concludeFull audit coverage does not cancel the findings raised inside the audited games.
INFO	simulation	Simulation produced no winner-direction intervention on this slate flip candidates 0; published flips 0	ATL@CHC, BAL@NYM, CHW@CLE, DET@TOR, LAD@CIN, MIA@ARI, NYY@MIN, SD@COL, SEA@LAA, SF@STL	Do thisMeasure the simulation's contribution across rolling slates, separating score realization from winner flipping. Don't concludeZero winner flips is not a defect and does not mean the simulation did no score-reconciliation work.

Finding	Persistence	Target lever(s)	Direction	Guardrail
simulation_inert_on_slate	11 consecutive · 11 in window	simulation.allow_score	confirm the simulation module is running, not silently no-oping	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.
weak_run_model_signal	11 consecutive · 11 in window	training.xgb.n_estimators, training.xgb.max_depth, training.xgb.learning_rate	investigate features/target via isolated training backtest (manual)	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.

Finding	Consecutive slates	Incidence	Present now	Fix type
high_raw_probability_cluster	0	1/11	no	param
pipeline_winner_instability	0	2/11	no	code/diagnostic
score_margin_compression	0	1/11	no	param
simulation_inert_on_slate	11	11/11	yes	code/diagnostic
single_slate_calibration_instability	1	5/11	yes	param
weak_run_model_signal	11	11/11	yes	code/diagnostic

Metric	Direction	Series (old → new)
winner_accuracy	improving	[0.625, 0.5333, 0.6, 0.5455, 0.7333, 0.6667, 0.6, 0.6667, 0.5333, 0.6, 0.7]
weighted_calibration_error	regressing	[0.1316, 0.1, 0.0937, 0.2611, 0.1825, 0.1069, 0.1171, 0.2091, 0.1711, 0.0827, 0.2812]
margin_bias	improving	[-0.688, -1.467, -1.467, -0.182, -0.867, -2.133, -1.0, -2.933, -2.667, -1.8, -0.4]
team_run_error_on_losses	regressing	[6.0, 5.286, 5.333, 3.8, 6.75, 5.6, 6.5, 5.0, 6.143, 5.0, 6.333]

Stage	Winner hits	Accuracy	Team-run MAE	Total-run MAE	Margin MAE
historical_ml	6/10	60%	2.083	3.343	1.881
variance_applied	5/10	50%	1.95	3.3	1.7
enhanced_prediction	7/10	70%	2.1	3.4	2
embedded_savant_run_environment	6/10	60%	2.15	3.3	2.3
fundamental_edge_target_review	7/10	70%	2.1	3.4	2.2
capped_target	7/10	70%	2.1	3.4	2.2
simulation_result	7/10	70%	2.1	3.4	2.2
published_final	7/10	70%	2.1	3.4	2.2

Transition	Δ accuracy	Δ hits	Δ team-run MAE	Δ margin MAE	Games whose winner/tie changed
historical_ml → variance_applied	-0.1	-1	-0.1328	-0.1811	DET@TOR: TOR → TIE; MIA@ARI: MIA → TIE; NYY@MIN: MIN → TIE; SF@STL: STL → TIE
variance_applied → enhanced_prediction	0.2	2	0.15	0.3	CHW@CLE: CHW → CLE; DET@TOR: TIE → DET; MIA@ARI: TIE → MIA; NYY@MIN: TIE → NYY; SF@STL: TIE → STL
enhanced_prediction → embedded_savant_run_environment	-0.1	-1	0.05	0.3	ATL@CHC: CHC → TIE
embedded_savant_run_environment → fundamental_edge_target_review	0.1	1	-0.05	-0.1	ATL@CHC: TIE → CHC
fundamental_edge_target_review → capped_target	0	0	0	0
capped_target → simulation_result	0	0	0	0
simulation_result → published_final	0	0	0	0

Game	Outcome	Raw model	Calibrated	Governor	Score coherence	Published	Explanation
ATL@CHC	hit	0.5538	0.535	0.535	0.535	0.5238	no recorded transformation explains the publication delta
BAL@NYM	miss	0.6845	0.725	0.725	0.725	0.6275	no recorded transformation explains the publication delta
CHW@CLE	miss	0.5	0.5078	0.5078	0.5078	0.5057	published probability matches the calibrated/governed value
DET@TOR	hit	0.5	0.5078	0.5078	0.5078	0.5057	published probability matches the calibrated/governed value
LAD@CIN	hit	0.7302	0.7259	0.7259	0.7259	0.58	score coherence followed by matchup-recency guardrail
MIA@ARI	miss	0.5242	0.5676	0.5676	0.5676	0.5174	no recorded transformation explains the publication delta
NYY@MIN	hit	0.5618	0.6446	0.6446	0.6446	0.57	score coherence followed by matchup-recency guardrail
SD@COL	hit	0.6914	0.6705	0.6705	0.6634	0.58	score coherence followed by matchup-recency guardrail
SEA@LAA	hit	0.62	0.5674	0.5674	0.5674	0.5276	score coherence followed by matchup-recency guardrail
SF@STL	hit	0.5143	0.5113	0.5113	0.5113	0.505	score coherence followed by matchup-recency guardrail

Sides analyzed	Broke on pitching	Broke on offense	Both	Variance	Mean starter ERA error	Mean lineup runs vs proj
20	2	6	1	11	-0.48	-0.2

Game	Trained p(home)	Blended p(home)	Blend favoured	Published p(home)	Published side	Artifact says winner changed	Outcome
CHW@CLE	0.462	0.485	away	0.506	home	yes	miss
DET@TOR	0.588	0.54	home	0.494	away	yes	hit

Slate	Games	Winner acc	Mean starter ERA error	Mean lineup runs vs proj	Sides broke on pitching	Sides broke on offense
20260913	15	0.600	-0.07	0.03	2	8
20260912	15	0.533	1.35	0.57	6	7
20260911	15	0.667	0.99	0.00	4	9
20260910	5	0.600	-3.19	-1.70		8
20260909	15	0.667	3.36	1.07	8	7
20260908	15	0.733	-0.15	-0.23	7	15
20260907	11	0.545	-0.86	-0.36	5	4
20260906	15	0.600	0.74	-0.30	6	10
20260905	15	0.533	-1.17	-0.07	2	6
20260904	16	0.625	-0.23	-0.44	5	12
20260903	9	0.556	0.24	-0.83	4	6
20260902	15	0.333	0.72	0.30	7	6

Game	Pred margin	Actual margin	Pred − actual
ATL@CHC	1	4	-3
BAL@NYM	3	1	2
CHW@CLE	1	4	-3
DET@TOR	1	1	0
LAD@CIN	5	3	2
MIA@ARI	1	1	0
NYY@MIN	2	5	-3
SD@COL	3	1	2
SEA@LAA	1	2	-1
SF@STL	1	1	0

Team	Starter	Projected	Actual	ER err
ATL	Reynaldo López	3 ER / 14 outs	5 ER / 9 outs	2
CHC	David Peterson	4 ER / 13 outs	1 ER / 18 outs	-3

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
ATL	Reynaldo López (Starter)	9 / 14	15 / 5.79 (+9.21)	5 / 3	6 / 5	0 / 3	2 / 2	1.8	2 / 4	4.57	venue away: 3.7 vs 5.1 ERA (ok) · opp run factor 1.08 (strong) · 25 prior appearances this season
CHC	David Peterson (Starter)	18 / 13	1.5 / 8.31 (-6.81)	1 / 4	3 / 6	0 / 0	3 / 3	3.33	7 / 6	4.95	venue home: 6.93 vs 4.5 ERA (worse here) · opp run factor 1.03 (strong) · 29 prior appearances this season
CHC	Ryan Zeferjahn (Setup Man)	3 / 0	0 / n/a	0 / 0	2 / 0	0 / 0	1 / 1	0.67	1 / 0	1.26

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
Matt Olson (1B)	ATL	3 / 4	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	1 / 1	3 / 1	-1.4	0.355 elite	bad day
Pete Crow-Armstrong (CF)	CHC	5 / 4	4 / 1	1 / 0	0 / 0	1 / 1	4 / 1	1 / 1	0 / 1	0 / 1	1.39	0.374 elite	projection too low
Seiya Suzuki (RF)	CHC	5 / 4	1 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	0 / 1	2 / 1	-1.26	0.341 elite	bad day
Mauricio Dubón (LF)	ATL	3 / 4	2 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	1 / 0	0 / 0	0.8	0.305 average	no signal
Michael Conforto (DH)	CHC	4 / 3	1 / 0	1 / 0	0 / 0	0 / 0	1 / 0	0 / 0	0 / 0	1 / 1	0.78	0.33 above average	projection too low
Drake Baldwin (DH)	ATL	3 / 4	2 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	1 / 1	1 / 1	0.47	0.362 elite	—
Ronald Acuña Jr. (RF)	ATL	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	1 / 1	1 / 1	-0.47	0.362 elite	—
Ozzie Albies (2B)	ATL	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	1 / 0	-0.47	0.281 weak	—
Sean Murphy (C)	ATL	4 / 3	0 / 1	0 / 0	0 / 0	0 / 0	1 / 0	1 / 0	0 / 0	2 / 1	-0.47	0.276 weak	—
Ha-Seong Kim (SS)	ATL	2 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	-0.47	0.229 weak	—
Alex Bregman (3B)	CHC	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	0 / 0	0 / 0	-0.47	0.321 above average	—
Nico Hoerner (SS)	CHC	4 / 4	2 / 1	0 / 0	0 / 0	0 / 0	1 / 0	2 / 0	0 / 0	0 / 0	0.47	0.324 above average	—
Michael Busch (1B)	CHC	2 / 4	0 / 1	0 / 1	0 / 0	0 / 0	0 / 1	1 / 1	2 / 1	1 / 1	-0.45	0.33 above average	—
Carson Kelly (C)	CHC	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	1 / 1	2 / 0	1 / 0	0 / 1	0.33	0.31 average	—
Ian Happ (LF)	CHC	3 / 4	1 / 1	1 / 0	0 / 0	0 / 0	0 / 0	1 / 0	1 / 1	0 / 1	0.31	0.316 average	—
Michael Harris II (CF)	ATL	4 / 4	2 / 1	0 / 1	0 / 0	0 / 0	0 / 1	1 / 0	0 / 0	0 / 1	0.16	0.34 elite	—
Pedro Ramirez (2B)	CHC	3 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	1 / 0	0 / 1	-0.14	0.29 below average	—
Austin Riley (3B)	ATL	4 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0	0.3 below average	—

Signal	Home-away edge	Meaning
Season quality	0.108	positive favors home; negative favors away
Recent form	0.011	positive favors home; negative favors away
Starting-pitcher matchup	0.049	positive favors home; negative favors away
Composite	0.169	positive favors home; negative favors away

Stage	Away-home runs	Confidence	What changed
historical_ml	4.39-4.81	n/a	Base ML historical model prediction
variance_applied	4-5	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	4-5	0.524	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by winner_starter_is_below_average, projected_margin_is_one_run)
embedded_savant_run_environment	5-5	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	5-6	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	5-6	n/a	Capped at max 12 realistic runs
simulation_result	5-6	n/a	Final inning-by-inning simulation result

Stage	Before	After	Toward final pick	What changed
variance_applied	4.39-4.81	4-5	0.578	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	4-5	4-5	0	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by winner_starter_is_below_average, projected_margin_is_one_run)
embedded_savant_run_environment	4-5	5-5	-1	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	5-5	5-6	1	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	5-6	5-6	0	Capped at max 12 realistic runs
simulation_result	5-6	5-6	0	Final inning-by-inning simulation result

Team	Player	Role	Net effect	Largest change
ATL	Bailey Falter	pitcher / Reliever	worse	predicted_H 5
ATL	Ronald Acuña Jr.	batter / RF	deboosted	predicted_AB -1.96
CHC	Pete Crow-Armstrong	batter / CF	worse	predicted_SO 1.845
ATL	Sean Murphy	batter / C	worse	predicted_SO 1.792
ATL	Ozzie Albies	batter / 2B	boosted	predicted_RBI 1.64
CHC	Pedro Ramirez	batter / 2B	better	predicted_RBI 1.54
CHC	Michael Conforto	batter / DH	better	predicted_AB 1.5
CHC	Ian Happ	batter / LF	deboosted	predicted_AB 1.414

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	ATL pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	ATL
info	CURRENT_PITCHER_DUPLICATION	CHC pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	CHC
info	CURRENT_FINALIZED_MATCH_BOX	ATL model-prediction batting totals differ from final simulation box score (AB: current=33, sim=37; H: current=9, sim=7; 2B: current=1, sim=2; R: current=4, sim=5; BB: current=3, sim=4; SO: current=7, sim=9).	ATL
info	CURRENT_PLAYER_MATCH_BOX	ATL current finalized hitter lines are model predictions and differ from the simulation box score: Drake Baldwin: RBI current=1 box=0; Michael Harris II: AB current=4 box=5, RBI current=1 box=0, SO current=1 box=2; Matt Olson: H current=1 box=0, HR current=1 box=0, RBI current=1 box=0, R current=1 box=0; Ronald Acuña Jr.: AB current=4 box=3, BB current=1 box=2, SO current=1 box=0; Ozzie Albies: AB current=4 box=5, H current=1 box=2, 2B current=0 box=1, RBI current=0 box=2; Austin Riley: AB current=3 box=4, HR current=0 box=1, RBI current=0 box=1, R current=0 box=1; Mauricio Dubón: H current=1 box=0, RBI current=1 box=0; Sean Murphy: AB current=3 box=4, RBI current=0 box=1, R current=0 box=1, SO current=1 box=3	ATL
info	CURRENT_FINALIZED_MATCH_BOX	CHC model-prediction batting totals differ from final simulation box score (AB: current=33, sim=40; H: current=8, sim=11; 2B: current=1, sim=3; HR: current=2, sim=3; RBI: current=5, sim=6; R: current=5, sim=6; BB: current=4, sim=3; HBP: current=0, sim=1).	CHC
info	CURRENT_PLAYER_MATCH_BOX	CHC current finalized hitter lines are model predictions and differ from the simulation box score: Alex Bregman: 2B current=0 box=1, RBI current=1 box=0, R current=1 box=0, HBP current=0 box=1; Michael Busch: H current=1 box=2, 2B current=1 box=0, HR current=0 box=1, SO current=1 box=2; Nico Hoerner: AB current=4 box=5; Seiya Suzuki: AB current=4 box=5, BB current=1 box=0, SO current=1 box=0; Pete Crow-Armstrong: AB current=4 box=5, BB current=1 box=0, SO current=1 box=3; Pedro Ramirez: AB current=3 box=4, RBI current=0 box=2, BB current=0 box=1; Ian Happ: AB current=4 box=5, H current=1 box=2, 2B current=0 box=1, R current=0 box=1; Carson Kelly: AB current=3 box=4, H current=1 box=0, RBI current=1 box=0	CHC
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.

Team	Starter	Projected	Actual	ER err
BAL	Brandon Young	3 ER / 16 outs	1 ER / 15 outs	-2
NYM	Jonah Tong	1 ER / 15 outs	0 ER / 18 outs	-1

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
BAL	Brandon Young (Starter)	15 / 16	1.8 / 5.06 (-3.26)	1 / 3	5 / 6	1 / 2	5 / 1	1.83	6 / 8	4.67	venue away: 3.39 vs 3.69 ERA (ok) · opp run factor 0.99 (average) · 25 prior appearances this season
BAL	Rico Garcia (Setup Man)	3 / 5	0 / 5.4 (-5.40)	0 / 1	0 / 2	0 / 1	0 / 0	0.25	0 / 2	1.27
BAL	Yennier Cano (Middle Reliever (High Leverage))	1 / 3	0 / 18 (-18.00)	0 / 2	0 / 3	0 / 1	0 / 3	0.16	0 / 2	0.71
NYM	Jonah Tong (Starter)	18 / 15	0 / 1.8	0 / 1	4 / 6	0 / 0	1 / 2	1.03	6 / 5	6.8	opp run factor 1 (average) · thin sample: 4 prior appearances this season (high uncertainty)
NYM	Kodai Senga (Closer)	3 / 3	0 / 9 (-9.00)	0 / 1	2 / 1	0 / 1	0 / 0	0.69	3 / 1	1.38

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
Pete Alonso (1B)	BAL	3 / 4	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	1 / 1	1 / 1	-1.4	0.395 elite	bad day
Heston Kjerstad (RF)	BAL	4 / 3	2 / 0	1 / 0	0 / 0	0 / 0	1 / 0	0 / 0	0 / 0	2 / 1	1.25	0.218 weak	good day
A.J. Ewing (CF)	NYM	3 / 4	1 / 1	0 / 0	0 / 0	1 / 0	1 / 0	1 / 1	1 / 1	0 / 1	0.93	0.322 above average	projection too low
Juan Soto (LF)	NYM	3 / 4	1 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	1 / 1	1 / 0	-0.93	0.413 elite	bad day
Dylan Beavers (LF)	BAL	5 / 4	0 / 1	0 / 1	0 / 0	0 / 0	0 / 1	0 / 0	0 / 0	2 / 1	-0.78	0.301 below average	projection too high
Gunnar Henderson (SS)	BAL	4 / 4	1 / 1	0 / 0	0 / 0	1 / 0	1 / 0	1 / 1	0 / 1	1 / 1	0.6	0.318 average	no signal
Colton Cowser (CF)	BAL	4 / 3	2 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0.47	0.316 average	—
Carson Benge (RF)	NYM	4 / 4	2 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 1	0.47	0.334 above average	—
Francisco Lindor (SS)	NYM	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	1 / 1	2 / 1	-0.47	0.333 above average	—
Bo Bichette (3B)	NYM	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 0	1 / 1	-0.47	0.329 above average	—
Jared Young (DH)	NYM	2 / 4	0 / 1	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	1 / 0	1 / 1	-0.45	0.314 average	—
Brett Baty (1B)	NYM	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	1 / 1	0.33	0.312 average	—
Christian Encarnacion-Strand (DH)	BAL	4 / 4	1 / 1	1 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 0	2 / 1	0.31	0.345 elite	—
Jackson Holliday (2B)	BAL	4 / 4	2 / 1	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0 / 1	0 / 1	0.14	0.293 below average	—
Coby Mayo (3B)	BAL	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	0 / 0	1 / 2	0	0.327 above average	—
Carlos Narváez (C)	BAL	1 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0	0.26 weak	—
Marcus Semien (2B)	NYM	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0	0.305 average	—
Francisco Alvarez (C)	NYM	4 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0	0.325 above average	—

Signal	Home-away edge	Meaning
Season quality	0.337	positive favors home; negative favors away
Recent form	1	positive favors home; negative favors away
Starting-pitcher matchup	-0.019	positive favors home; negative favors away
Composite	0.934	positive favors home; negative favors away

Stage	Away-home runs	Confidence	What changed
historical_ml	4.27-4.61	n/a	Base ML historical model prediction
variance_applied	4-5	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	4-7	0.628	Enhanced ML with recent form, H2H, low-score analysis
embedded_savant_run_environment	4-7	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	4-7	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	4-7	n/a	Capped at max 12 realistic runs
simulation_result	4-7	n/a	Final inning-by-inning simulation result

Stage	Before	After	Toward final pick	What changed
variance_applied	4.27-4.61	4-5	0.667	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	4-5	4-7	2	Enhanced ML with recent form, H2H, low-score analysis
embedded_savant_run_environment	4-7	4-7	0	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	4-7	4-7	0	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	4-7	4-7	0	Capped at max 12 realistic runs
simulation_result	4-7	4-7	0	Final inning-by-inning simulation result

Team	Player	Role	Net effect	Largest change
BAL	Colton Cowser	batter / CF	better	predicted_H 2.46
NYM	Brett Baty	batter / 1B	better	predicted_H 2.27
NYM	Francisco Lindor	batter / SS	better	predicted_RBI 2.19
NYM	Carson Benge	batter / RF	worse	predicted_SO 2.065
BAL	Brandon Young	pitcher / Starter	neutral	predicted_SO 2
NYM	Jared Young	batter / DH	worse	predicted_AB 1.828
BAL	Heston Kjerstad	batter / RF	better	predicted_AB 1.5
BAL	Dylan Beavers	batter / LF	boosted	predicted_RBI 1.43

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	BAL pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	BAL
info	CURRENT_PITCHER_DUPLICATION	NYM pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	NYM
info	CURRENT_FINALIZED_MATCH_BOX	BAL model pitcher projections differ from final simulation pitching box score (H: current=11, sim=12; ER: current=6, sim=7).	BAL
info	CURRENT_FINALIZED_MATCH_BOX	BAL model-prediction batting totals differ from final simulation box score (AB: current=33, sim=36; H: current=7, sim=9; 2B: current=1, sim=0; HR: current=1, sim=2; BB: current=3, sim=2; SO: current=10, sim=6; HBP: current=0, sim=2).	BAL
info	CURRENT_PLAYER_MATCH_BOX	BAL current finalized hitter lines are model predictions and differ from the simulation box score: Dylan Beavers: AB current=4 box=5, H current=1 box=2, 2B current=1 box=0, HR current=0 box=1; Gunnar Henderson: H current=1 box=0, R current=1 box=0; Christian Encarnacion-Strand: RBI current=1 box=0, SO current=1 box=0; Heston Kjerstad: AB current=3 box=4, SO current=1 box=0; Coby Mayo: RBI current=1 box=0, SO current=2 box=1, HBP current=0 box=1; Pete Alonso: AB current=4 box=5, H current=1 box=0, HR current=1 box=0, RBI current=1 box=0; Jackson Holliday: H current=1 box=2, BB current=1 box=0, SO current=1 box=0; Carlos Narváez: BB current=0 box=1, SO current=1 box=2	BAL
info	CURRENT_FINALIZED_MATCH_BOX	NYM model-prediction batting totals differ from final simulation box score (AB: current=34, sim=36; H: current=8, sim=12; 2B: current=1, sim=2; HR: current=1, sim=4; RBI: current=4, sim=7; R: current=5, sim=7; BB: current=3, sim=4; SO: current=8, sim=12).	NYM
info	CURRENT_PLAYER_MATCH_BOX	NYM current finalized hitter lines are model predictions and differ from the simulation box score: Carson Benge: AB current=4 box=5, SO current=1 box=3; Jared Young: AB current=4 box=5, H current=1 box=0, 2B current=1 box=0, RBI current=1 box=0; Francisco Lindor: H current=1 box=3, HR current=0 box=2, RBI current=1 box=3, SO current=1 box=0; A.J. Ewing: AB current=4 box=3, HR current=0 box=1, RBI current=0 box=1, BB current=1 box=0; Brett Baty: AB current=3 box=4, H current=0 box=3, 2B current=0 box=1, R current=0 box=1; Juan Soto: H current=1 box=2, SO current=0 box=1; Marcus Semien: AB current=4 box=3, H current=1 box=0, BB current=0 box=1, SO current=1 box=2; Bo Bichette: R current=0 box=1, BB current=0 box=1	NYM
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.

Team	Starter	Projected	Actual	ER err
CHW	Sean Newcomb	3 ER / 15 outs	0 ER / 3 outs	-3
CLE	Gavin Williams	1 ER / 16 outs	2 ER / 15 outs	1

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
CLE	Gavin Williams (Starter)	15 / 16	3.6 / 1.69	2 / 1	4 / 2	0 / 0	3 / 4	2.14	7 / 7	7.33	venue home: 3.01 vs 4.79 ERA (ok) · opp run factor 0.96 (average) · 29 prior appearances this season
CLE	Tim Herrin (Middle Reliever (High Leverage))	2 / 2	13.5 / 27 (-13.50)	1 / 2	3 / 3	0 / 0	0 / 1	0.41	1 / 0	0.85

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
Munetaka Murakami (1B)	CHW	3 / 4	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	1 / 1	3 / 2	-1.4	0.342 elite	bad day
Miguel Vargas (3B)	CHW	4 / 4	2 / 1	0 / 0	0 / 0	1 / 0	2 / 1	2 / 1	1 / 1	0 / 1	1.4	0.379 elite	projection too low
Angel Martínez (LF)	CLE	3 / 4	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	0 / 0	1 / 1	-1.4	0.306 average	no signal
José Ramírez (3B)	CLE	4 / 4	1 / 1	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	1 / 1	-0.64	0.326 above average	bad day
Chase DeLauter (DH)	CLE	3 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	1 / 1	0 / 1	-0.47	0.344 elite	—
Patrick Bailey (C)	CLE	4 / 3	1 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	0 / 0	2 / 1	0.47	0.308 average	—
Nathaniel Lowe (1B)	CLE	2 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 0	1 / 1	-0.47	0.338 above average	—
Travis Bazzana (2B)	CLE	4 / 3	1 / 1	1 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 1	0.31	0.307 average	—
Jo Adell (RF)	CLE	4 / 4	1 / 1	0 / 1	0 / 0	0 / 0	0 / 1	1 / 1	0 / 0	1 / 1	-0.31	0.325 above average	—
Braden Montgomery (RF)	CHW	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	3 / 1	0	0.331 above average	—
Colson Montgomery (SS)	CHW	4 / 4	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	0 / 0	3 / 2	0	0.304 average	—
Chase Meidroth (2B)	CHW	4 / 3	1 / 1	0 / 0	0 / 0	0 / 0	1 / 0	1 / 1	0 / 0	1 / 1	0	0.296 below average	—
Steven Kwan (CF)	CLE	3 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	1 / 1	0 / 0	0	0.322 above average	—
Brayan Rocchio (SS)	CLE	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0	0.293 below average	—

Signal	Home-away edge	Meaning
Season quality	-0.003	positive favors home; negative favors away
Recent form	0.302	positive favors home; negative favors away
Starting-pitcher matchup	0.03	positive favors home; negative favors away
Composite	0.321	positive favors home; negative favors away

Stage	Away-home runs	Confidence	What changed
historical_ml	4.55-4.38	n/a	Base ML historical model prediction
variance_applied	5-4	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	4-5	0.506	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by projected_margin_is_one_run, matchup_recency_guardrail)
embedded_savant_run_environment	4-5	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	4-5	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	4-5	n/a	Capped at max 12 realistic runs
simulation_result	4-5	n/a	Final inning-by-inning simulation result

Stage	Before	After	Toward final pick	What changed
variance_applied	4.55-4.38	5-4	-0.825	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	5-4	4-5	2	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by projected_margin_is_one_run, matchup_recency_guardrail)
embedded_savant_run_environment	4-5	4-5	0	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	4-5	4-5	0	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	4-5	4-5	0	Capped at max 12 realistic runs
simulation_result	4-5	4-5	0	Final inning-by-inning simulation result

Team	Player	Role	Net effect	Largest change
CLE	Gavin Williams	pitcher / Starter	neutral	predicted_SO 5
CHW	Sam Antonacci	batter / LF	better	predicted_RBI 2.656
CLE	Erik Sabrowski	pitcher / Middle Reliever (High Leverage)	worse	predicted_SO -2
CLE	Tim Herrin	pitcher / Middle Reliever (High Leverage)	worse	predicted_SO -2
CHW	Kyle Teel	batter / C	better	predicted_2B 1.929
CHW	Miguel Vargas	batter / 3B	worse	predicted_AB -1.79
CHW	Colson Montgomery	batter / SS	worse	predicted_SO -1.461
CLE	Chase DeLauter	batter / DH	worse	predicted_SO 1.37

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	CHW pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	CHW
info	CURRENT_PITCHER_DUPLICATION	CLE pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	CLE
info	CURRENT_FINALIZED_MATCH_BOX	CHW model-prediction batting totals differ from final simulation box score (AB: current=32, sim=35; H: current=7, sim=8; 2B: current=2, sim=3; HR: current=1, sim=0; BB: current=4, sim=5; SO: current=11, sim=7).	CHW
info	CURRENT_PLAYER_MATCH_BOX	CHW current finalized hitter lines are model predictions and differ from the simulation box score: Sam Antonacci: AB current=4 box=5, H current=1 box=2, 2B current=0 box=1, RBI current=0 box=3; Munetaka Murakami: H current=1 box=0, HR current=1 box=0, RBI current=1 box=0, R current=1 box=0; Kyle Teel: AB current=4 box=5, H current=1 box=2, 2B current=0 box=2, R current=0 box=1; Andrew Benintendi: AB current=3 box=4, H current=1 box=0, 2B current=1 box=0, RBI current=1 box=0; Tristan Peters: AB current=3 box=4, H current=1 box=0, 2B current=1 box=0, SO current=1 box=0; Miguel Vargas: AB current=4 box=3, RBI current=1 box=0, BB current=1 box=2; Colson Montgomery: AB current=4 box=3, H current=0 box=2, R current=0 box=1, BB current=0 box=1; Chase Meidroth: AB current=3 box=4, R current=1 box=0	CHW
info	CURRENT_FINALIZED_MATCH_BOX	CLE model-prediction batting totals differ from final simulation box score (AB: current=33, sim=32; 2B: current=2, sim=0; RBI: current=4, sim=5; BB: current=3, sim=0; SO: current=7, sim=6).	CLE
info	CURRENT_PLAYER_MATCH_BOX	CLE current finalized hitter lines are model predictions and differ from the simulation box score: Steven Kwan: H current=1 box=0, R current=1 box=0, BB current=1 box=0; Chase DeLauter: RBI current=1 box=0, R current=0 box=1, BB current=1 box=0, SO current=1 box=2; Travis Bazzana: AB current=3 box=4, RBI current=0 box=2; José Ramírez: H current=1 box=2, 2B current=1 box=0, RBI current=0 box=1, BB current=1 box=0; Angel Martínez: AB current=4 box=3, H current=1 box=0, HR current=1 box=0, RBI current=1 box=0; Jo Adell: H current=1 box=2, 2B current=1 box=0, HR current=0 box=1, RBI current=1 box=2; Patrick Bailey: H current=0 box=1, R current=0 box=1, SO current=1 box=0; Nathaniel Lowe: AB current=4 box=3, H current=1 box=0, RBI current=1 box=0	CLE
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.

Team	Starter	Projected	Actual	ER err
DET	Troy Melton	3 ER / 18 outs	2 ER / 8 outs	-1
TOR	José Soriano	5 ER / 17 outs	4 ER / 19 outs	-1

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
DET	Troy Melton (Starter)	8 / 18	6.75 / 4.5	2 / 3	5 / 7	0 / 1	4 / 2	1.47	3 / 6	4.16	venue away: 2.06 vs 2.81 ERA (ok) · opp run factor 0.93 (weak) · 18 prior appearances this season
DET	Tyler Holton (Setup Man)	4 / 4	0 / 0	0 / 0	0 / 1	0 / 0	0 / 0	0.38	0 / 2	1.09
DET	Kenley Jansen (Closer)	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0.48	1 / 1	1.16
DET	Kyle Finnegan (Middle Reliever (High Leverage))	3 / 2	27 / 13.5 (+13.50)	3 / 1	3 / 1	1 / 0	0 / 0	0.55	1 / 2	0.86
TOR	José Soriano (Starter)	19 / 17	5.68 / 7.94	4 / 5	5 / 4	0 / 1	1 / 4	1.78	5 / 7	3.61	venue home: 3.61 vs 3.39 ERA (ok) · opp run factor 1 (average) · 29 prior appearances this season
TOR	Louis Varland (Closer)	5 / 3	5.4 / 0 (+5.40)	1 / 0	2 / 0	0 / 0	1 / 0	0.32	2 / 2	1.05

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
Dillon Dingler (C)	DET	4 / 4	1 / 1	0 / 0	0 / 0	0 / 1	1 / 1	1 / 1	0 / 0	2 / 1	-0.93	0.339 above average	bad day
Spencer Torkelson (1B)	DET	3 / 3	1 / 0	1 / 0	0 / 0	0 / 0	1 / 0	0 / 0	1 / 1	2 / 1	0.78	0.309 average	no signal
Hao-Yu Lee (3B)	DET	4 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	-0.47	0.298 below average	—
Alejandro Kirk (C)	TOR	4 / 4	2 / 1	0 / 0	0 / 0	0 / 0	1 / 1	1 / 0	1 / 1	0 / 1	0.47	0.308 average	—
George Springer (DH)	TOR	2 / 4	1 / 1	0 / 1	0 / 0	0 / 0	0 / 1	2 / 1	3 / 1	0 / 1	0.35	0.306 average	—
Gleyber Torres (2B)	DET	5 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 1	1 / 1	0 / 1	1 / 1	-0.33	0.325 above average	—
Zach McKinstry (RF)	DET	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	1 / 0	0 / 1	0.33	0.285 below average	—
Brett Bateman (CF)	TOR	5 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0 / 1	1 / 1	-0.33	0.312 average	—
Josh Smith (LF)	TOR	1 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	1 / 1	0.33	0.284 weak	—
Max Clark (CF)	DET	4 / 3	1 / 1	1 / 0	0 / 0	0 / 0	2 / 0	1 / 1	0 / 0	0 / 1	0.31	0.323 above average	—
Nathan Lukes (RF)	TOR	4 / 4	1 / 1	1 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0.31	0.287 below average	—
Riley Greene (DH)	DET	5 / 4	2 / 1	0 / 1	0 / 0	0 / 0	0 / 1	1 / 1	0 / 1	1 / 1	-0.17	0.372 elite	—
Kevin McGonigle (SS)	DET	5 / 4	2 / 1	0 / 0	0 / 0	0 / 0	0 / 1	1 / 1	0 / 1	0 / 0	0.14	0.354 elite	—
Vladimir Guerrero Jr. (1B)	TOR	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	1 / 0	1 / 0	-0.14	0.334 above average	—
Brett Callahan (LF)	DET	4 / 3	1 / 1	0 / 0	0 / 0	0 / 0	1 / 0	0 / 0	0 / 0	0 / 1	0	0.299 below average	—
Andrés Giménez (SS)	TOR	4 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	2 / 1	0	0.274 weak	—
Ernie Clement (2B)	TOR	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0	0.261 weak	—
Kazuma Okamoto (3B)	TOR	4 / 3	1 / 1	0 / 0	0 / 0	1 / 1	3 / 1	1 / 1	0 / 0	1 / 1	0	0.329 above average	—

Signal	Home-away edge	Meaning
Season quality	-0.464	positive favors home; negative favors away
Recent form	-0.035	positive favors home; negative favors away
Starting-pitcher matchup	-0.064	positive favors home; negative favors away
Composite	-0.584	positive favors home; negative favors away

Stage	Away-home runs	Confidence	What changed
historical_ml	4.1-4.31	n/a	Base ML historical model prediction
variance_applied	4-4	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	5-4	0.506	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by projected_margin_is_one_run, matchup_recency_guardrail)
embedded_savant_run_environment	5-4	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	5-4	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	5-4	n/a	Capped at max 12 realistic runs
simulation_result	5-4	n/a	Final inning-by-inning simulation result

Stage	Before	After	Toward final pick	What changed
variance_applied	4.1-4.31	4-4	0.21	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	4-4	5-4	1	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by projected_margin_is_one_run, matchup_recency_guardrail)
embedded_savant_run_environment	5-4	5-4	-0	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	5-4	5-4	-0	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	5-4	5-4	-0	Capped at max 12 realistic runs
simulation_result	5-4	5-4	-0	Final inning-by-inning simulation result

Team	Player	Role	Net effect	Largest change
TOR	Alejandro Kirk	batter / C	worse	predicted_SO 3.485
TOR	José Soriano	pitcher / Starter	neutral	predicted_SO 3
DET	Gleyber Torres	batter / 2B	worse	predicted_SO 2.163
DET	Dillon Dingler	batter / C	worse	predicted_AB -2.02
TOR	George Springer	batter / DH	worse	predicted_H 1.87
TOR	Vladimir Guerrero Jr.	batter / 1B	worse	predicted_BB 1.627
DET	Brett Callahan	batter / LF	worse	predicted_AB 1.5
TOR	Brett Bateman	batter / CF	boosted	predicted_RBI 1.48

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	DET pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	DET
info	CURRENT_PITCHER_DUPLICATION	TOR pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	TOR
info	CURRENT_FINALIZED_MATCH_BOX	DET model-prediction batting totals differ from final simulation box score (AB: current=31, sim=34; HR: current=1, sim=2; RBI: current=4, sim=5; BB: current=4, sim=5; SO: current=8, sim=12).	DET
info	CURRENT_PLAYER_MATCH_BOX	DET current finalized hitter lines are model predictions and differ from the simulation box score: Riley Greene: AB current=4 box=5, H current=1 box=2, 2B current=1 box=0, BB current=1 box=0; Kevin McGonigle: H current=1 box=0, SO current=0 box=2; Dillon Dingler: AB current=4 box=3, BB current=0 box=1; Brett Callahan: AB current=3 box=4, H current=1 box=0, SO current=1 box=2; Gleyber Torres: H current=1 box=0, SO current=1 box=3; Hao-Yu Lee: H current=1 box=0, BB current=0 box=1, SO current=1 box=0; Max Clark: HR current=0 box=1, RBI current=0 box=1, BB current=0 box=1; Spencer Torkelson: AB current=3 box=4, H current=0 box=1, 2B current=0 box=1, BB current=1 box=0	DET
info	CURRENT_FINALIZED_MATCH_BOX	TOR model pitcher projections differ from final simulation pitching box score (H: current=6, sim=7; HR: current=1, sim=2).	TOR
info	CURRENT_FINALIZED_MATCH_BOX	TOR model-prediction batting totals differ from final simulation box score (AB: current=33, sim=36; H: current=8, sim=9; 2B: current=1, sim=2; SO: current=7, sim=11; HBP: current=0, sim=1).	TOR
info	CURRENT_PLAYER_MATCH_BOX	TOR current finalized hitter lines are model predictions and differ from the simulation box score: Alejandro Kirk: AB current=4 box=5, H current=1 box=0, RBI current=1 box=0, BB current=1 box=0; Brett Bateman: AB current=4 box=5, H current=1 box=2, RBI current=0 box=2, BB current=1 box=0; Vladimir Guerrero Jr.: AB current=4 box=3, H current=1 box=0, RBI current=1 box=0, R current=1 box=0; George Springer: AB current=4 box=5, H current=1 box=3, 2B current=1 box=2, RBI current=1 box=0; Josh Smith: H current=0 box=1, R current=0 box=1, BB current=0 box=1; Nathan Lukes: H current=1 box=0; Andrés Giménez: AB current=3 box=4, SO current=1 box=0; Ernie Clement: AB current=4 box=3, H current=1 box=0, HBP current=0 box=1	TOR
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.

Team	Starter	Projected	Actual	ER err
LAD	Tarik Skubal	3 ER / 20 outs	0 ER / 21 outs	-3
CIN	Nick Lodolo	7 ER / 13 outs	4 ER / 21 outs	-3

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
LAD	Tarik Skubal (Starter)	21 / 20	0 / 4.05 (-4.05)	0 / 3	3 / 5	0 / 1	1 / 1	1.15	9 / 7	6.66	venue away: 2.09 vs 3.4 ERA (ok) · opp run factor 0.94 (weak) · 23 prior appearances this season
CIN	Nick Lodolo (Starter)	21 / 13	5.14 / 14.54 (-9.40)	4 / 7	3 / 9	2 / 1	4 / 3	2.57	4 / 3	4.34	venue home: 4.5 vs 5.97 ERA (ok) · opp run factor 0.97 (strong) · 18 prior appearances this season
CIN	Luis Mey (Setup Man)	3 / 3	0 / 0	0 / 0	0 / 1	0 / 0	0 / 0	0.8	1 / 0	1.68
CIN	Tony Santillan (Middle Reliever (High Leverage))	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	2 / 1	0.75	1 / 2	1.01

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
Max Muncy (3B)	LAD	4 / 3	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	0 / 1	3 / 1	-1.73	0.35 elite	bad day
Kyle Tucker (RF)	LAD	2 / 4	1 / 1	0 / 0	0 / 0	1 / 0	1 / 0	2 / 1	2 / 1	0 / 1	1.26	0.327 above average	projection too low
Teoscar Hernández (LF)	LAD	3 / 4	1 / 1	0 / 0	0 / 0	1 / 0	2 / 1	2 / 0	1 / 0	1 / 1	1.26	0.331 above average	projection too low
Mookie Betts (SS)	LAD	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	0 / 0	-0.8	0.335 above average	bad day
Will Smith (C)	LAD	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	1 / 1	-0.8	0.382 elite	bad day
Tyler Stephenson (C)	CIN	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 1	2 / 1	-0.8	0.333 above average	bad day
Miguel Rojas (2B)	LAD	3 / 3	0 / 1	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	-0.78	0.304 average	no signal
Dane Myers (CF)	CIN	3 / 4	1 / 1	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	-0.64	0.31 average	no signal
Tommy Edman (CF)	LAD	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	0 / 0	1 / 1	-0.47	0.311 average	—
Sal Stewart (1B)	CIN	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	0 / 0	2 / 1	-0.47	0.35 elite	—
Enrique Hernández (1B)	LAD	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	1 / 0	0 / 0	1 / 0	0 / 1	0.33	0.275 weak	—
Josue De Paula (DH)	LAD	2 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	1 / 0	0 / 1	0.33	0.542 elite	—
Elly De La Cruz (SS)	CIN	4 / 4	1 / 1	0 / 0	0 / 0	1 / 1	1 / 1	1 / 1	0 / 1	2 / 2	-0.33	0.373 elite	—
JJ Bleday (LF)	CIN	2 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	1 / 1	-0.14	0.342 elite	—
Juan Brito (2B)	CIN	3 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 1	0	0.292 below average	—
Ke'Bryan Hayes (3B)	CIN	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	2 / 1	0	0.298 below average	—
Eugenio Suárez (RF)	CIN	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	0 / 0	0 / 2	0	0.299 below average	—

Signal	Home-away edge	Meaning
Season quality	-1.894	positive favors home; negative favors away
Recent form	-1.462	positive favors home; negative favors away
Starting-pitcher matchup	-0.029	positive favors home; negative favors away
Composite	-2.371	positive favors home; negative favors away

Stage	Away-home runs	Confidence	What changed
historical_ml	5.08-2.5	n/a	Base ML historical model prediction
variance_applied	5-2	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	7-3	0.58	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by matchup_recency_guardrail)
embedded_savant_run_environment	8-3	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	8-3	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	8-3	n/a	Capped at max 12 realistic runs
simulation_result	8-3	n/a	Final inning-by-inning simulation result

Stage	Before	After	Toward final pick	What changed
variance_applied	5.08-2.5	5-2	0.417	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	5-2	7-3	1	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by matchup_recency_guardrail)
embedded_savant_run_environment	7-3	8-3	1	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	8-3	8-3	-0	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	8-3	8-3	-0	Capped at max 12 realistic runs
simulation_result	8-3	8-3	-0	Final inning-by-inning simulation result

SD@COLHIT

Predicted: SD 8-5 (conf -0.16, moderate)Actual: SD 8-7Brier 0.176total bias -2

Root cause: Strong correct call

Evidence-backed diagnosis

published system edge 0.16 (moderate); this is not the half-scale distance from 50%
team-run absolute error 2; total bias -2
pregame home-minus-away evidence: season -1.366, recent -1.246, pitcher -0.035, composite -1.679
confidence pipeline reached 0.670 before later publication/recency controls; published home probability was 0.420
score-volatility actions: supported_blowout_margin_expansion
forecast-file integrity: clean (0 defects, 8 informational note(s)); the saved artifact is not why the pick missed

Quantified contributing factors

Tomoyuki Sugano (COL) allowed 3 more ER than projected (6 vs 3)

Team	Starter	Projected	Actual	ER err
SD	Casey Mize	3 ER / 15 outs	2 ER / 15 outs	-1
COL	Tomoyuki Sugano	3 ER / 14 outs	6 ER / 15 outs	3

Decision internals: run factors home×0.934 / away×0.968; savant lineup adjustment applied

Player-level breakdown (projected vs actual)

Every value is actual / projected. Run impact approximates how far each hitter's real line moved run expectancy from its projection; it orders the movers and is not a published stat.

Top deviations that drove the game:

Cole Carrigg (COL CF): 2-for-5, 1 HR, 1 RBI (projected 1.0 H, 0.0 HR) [run impact 1.09]
Jake McCarthy (COL LF): 2-for-5, 0 HR, 2 RBI (projected 1.0 H, 0.0 HR) [run impact 1.09]
Luis Campusano (SD DH): 0-for-5, 0 HR, 0 RBI (projected 1.0 H, 0.0 HR) [run impact -0.8]
Jake Cronenworth (SD 2B): 2-for-3, 0 HR, 1 RBI (projected 1.0 H, 0.0 HR) [run impact 0.8]
Hunter Goodman (COL C): 2-for-3, 0 HR, 0 RBI (projected 1.0 H, 1.0 HR) [run impact 0.8]

Starter miss: Tomoyuki Sugano (COL) posted 10.80 ERA vs projected 5.79 (6 ER in 15 outs); no credible venue, opponent-strength or experience signal explains it — an outlier start (variance).

What to do: Variance — no model change is warranted for this start.

Pitcher versus batter: what actually broke this game

Per-pitch matchup fields are not populated in the current feeds, so this is reconstructed from the settled box: each starter's workload-adjusted ERA, the bullpen share behind him, and the opposing lineup's real production against its projection, with the hitters who beat that projection named.

AWAY pitching

Casey Mize (SD) threw 15 outs, 2 ER = 3.6 ERA against a projected 5.4 ERA (-1.8 ERA error)

Bullpen behind him: 2 ER over 11 outs.

The HOME lineup he faced produced 7 runs on 14 hits and 1 HR, against a projected 4 runs / 9 hits (3 runs vs projection).

Who did the damage to him:

Cole Carrigg: 2-for-5, 1 HR, 1 RBI [+1.09 runs vs projection] season xwOBA 0.311, average - a good day, not a model error
Jake McCarthy: 2-for-5, 0 HR, 2 RBI [+1.09 runs vs projection] season xwOBA 0.321, above average - real projection error
Hunter Goodman: 2-for-3, 0 HR, 0 RBI [+0.8 runs vs projection] season xwOBA 0.318, average - a good day, not a model error

Verdict: neither side broke materially - the result is closer to variance

HOME pitching

Tomoyuki Sugano (COL) threw 15 outs, 6 ER = 10.8 ERA against a projected 5.79 ERA (5.01 ERA error)

Bullpen behind him: 0 ER over 3 outs.

The AWAY lineup he faced produced 8 runs on 13 hits and 2 HR, against a projected 5 runs / 9 hits (3 runs vs projection).

Who did the damage to him:

Jake Cronenworth: 2-for-3, 0 HR, 1 RBI [+0.8 runs vs projection] season xwOBA 0.268, weak - a good day, not a model error
Dustin Harris: 1-for-4, 0 HR, 1 RBI [+0.64 runs vs projection] season xwOBA 0.321, above average - real projection error
Jackson Merrill: 3-for-5, 0 HR, 1 RBI [+0.63 runs vs projection] season xwOBA 0.347, elite - real projection error

Verdict: pitching: the starter gave up runs at a far worse rate than projected

Starting pitchers & relievers: actual / projected + context (ERA is workload-aware: ER over 9 innings, so a short blow-up is not read like a quality start; ERA error in parentheses when material. exp BB and exp SO are the continuous rate expectations. The integer beside each is a simulation draw reconciled to the opposing simulated batting box, so it is not the line that ranks pitchers: on the 2026-09-07 slate the integers published 39 walks and 121 strikeouts against 51 and 130 real, while these expectations landed on 50.4 and 132.6.)

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
SD	Casey Mize (Starter)	15 / 15	3.6 / 5.4	2 / 3	8 / 3	0 / 1	2 / 0	1.13	0 / 4	3.58	venue away: 4.25 vs 3.15 ERA (worse here) · opp run factor 0.93 (weak) · 23 prior appearances this season
SD	Adrian Morejon (Setup Man)	3 / 3	9 / 0 (+9.00)	1 / 0	2 / 0	0 / 0	0 / 0	0.28	0 / 2	1.21
SD	Bradgley Rodriguez (Middle Reliever (High Leverage))	5 / 3	5.4 / 9 (-3.60)	1 / 1	2 / 2	1 / 0	0 / 0	0.21	0 / 1	1.19
SD	Mason Miller (Closer)	3 / 3	0 / 9 (-9.00)	0 / 1	1 / 3	0 / 0	0 / 1	0.4	2 / 2	1.83
COL	Tomoyuki Sugano (Starter)	15 / 14	10.8 / 5.79 (+5.01)	6 / 3	8 / 6	2 / 1	1 / 1	1.34	2 / 5	3.52	venue home: 4.96 vs 5.64 ERA (ok) · opp run factor 0.97 (average) · 26 prior appearances this season
COL	Jordan Romano (Closer)	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0.55	2 / 3	1.06

Hitters: actual / projected (sorted by run impact vs projection)

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
Cole Carrigg (CF)	COL	5 / 4	2 / 1	0 / 1	0 / 0	1 / 0	1 / 1	1 / 1	0 / 0	1 / 1	1.09	0.311 average	no signal
Jake McCarthy (LF)	COL	5 / 4	2 / 1	1 / 1	1 / 0	0 / 0	2 / 1	1 / 1	0 / 0	0 / 1	1.09	0.321 above average	projection too low
Luis Campusano (DH)	SD	5 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 1	2 / 1	-0.8	0.337 above average	bad day
Jake Cronenworth (2B)	SD	3 / 3	2 / 1	0 / 0	0 / 0	0 / 0	1 / 0	1 / 0	1 / 0	1 / 1	0.8	0.268 weak	good day
Hunter Goodman (C)	COL	3 / 4	2 / 1	1 / 0	1 / 0	0 / 1	0 / 1	1 / 1	1 / 0	0 / 1	0.8	0.318 average	no signal
Kyle Karros (3B)	COL	5 / 4	2 / 1	1 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0.78	0.337 above average	projection too low
Connor Norby (2B)	COL	5 / 4	2 / 1	2 / 0	0 / 0	0 / 0	2 / 0	1 / 0	0 / 1	0 / 1	0.76	0.305 average	no signal
Dustin Harris (LF)	SD	4 / 3	1 / 1	1 / 0	0 / 0	0 / 0	1 / 0	1 / 1	1 / 0	0 / 0	0.64	0.321 above average	projection too low
Jackson Merrill (CF)	SD	5 / 4	3 / 1	0 / 1	0 / 0	0 / 0	1 / 1	1 / 1	0 / 0	1 / 1	0.63	0.347 elite	projection too low
Fernando Tatis Jr. (RF)	SD	5 / 4	1 / 1	0 / 1	0 / 0	1 / 0	2 / 1	1 / 1	0 / 0	1 / 1	0.62	0.376 elite	projection too low
Manny Machado (3B)	SD	5 / 4	1 / 1	0 / 0	0 / 0	1 / 0	1 / 1	1 / 1	0 / 1	0 / 1	0.6	0.337 above average	projection too low
Adael Amador (DH)	COL	4 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 1	1 / 1	1 / 0	0 / 1	0.33	0.281 weak	—
Ezequiel Tovar (SS)	COL	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	2 / 0	1 / 0	0 / 1	0.33	0.279 weak	—
Ty France (1B)	SD	5 / 4	3 / 1	1 / 0	0 / 0	0 / 1	1 / 1	1 / 1	0 / 0	1 / 1	0.32	0.335 above average	—
Mickey Moniak (RF)	COL	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	1 / 0	1 / 1	-0.14	0.278 weak	—
TJ Rumfield (1B)	COL	5 / 4	2 / 1	0 / 0	0 / 0	0 / 0	2 / 0	0 / 0	0 / 1	0 / 0	0.14	0.335 above average	—
Xander Bogaerts (SS)	SD	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	2 / 0	1 / 1	0 / 0	0	0.318 average	—
Freddy Fermin (C)	SD	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	1 / 0	0 / 0	0 / 0	2 / 0	0	0.275 weak	—

Full model trace, player evidence, and audit findings

Fundamental evidence entering the pick

Signal	Home-away edge	Meaning
Season quality	-1.366	positive favors home; negative favors away
Recent form	-1.246	positive favors home; negative favors away
Starting-pitcher matchup	-0.035	positive favors home; negative favors away
Composite	-1.679	positive favors home; negative favors away

Score and confidence evolution

Stage	Away-home runs	Confidence	What changed
historical_ml	5.79-3.61	n/a	Base ML historical model prediction
variance_applied	6-4	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	7-5	0.58	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by matchup_recency_guardrail)
embedded_savant_run_environment	8-5	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	8-5	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	8-5	n/a	Capped at max 12 realistic runs
simulation_result	8-5	n/a	Final inning-by-inning simulation result

Official postgame stage attribution

Stage	Before	After	Toward final pick	What changed
variance_applied	5.79-3.61	6-4	-0.172	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	6-4	7-5	-0	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by matchup_recency_guardrail)
embedded_savant_run_environment	7-5	8-5	1	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	8-5	8-5	-0	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	8-5	8-5	-0	Capped at max 12 realistic runs
simulation_result	8-5	8-5	-0	Final inning-by-inning simulation result

Largest player-level adjustments from the ultra-detailed trace

These are pregame model adjustments, not claims about actual postgame player error.

Team	Player	Role	Net effect	Largest change
SD	Jake Cronenworth	batter / 2B	better	predicted_RBI 2.72
SD	Manny Machado	batter / 3B	better	predicted_RBI 2.18
COL	Connor Norby	batter / 2B	worse	predicted_SO 1.914
SD	Dustin Harris	batter / LF	better	predicted_AB 1.88
COL	Cole Carrigg	batter / CF	deboosted	predicted_RBI 1.494
COL	Jake McCarthy	batter / LF	worse	predicted_RBI 1.33
COL	Hunter Goodman	batter / C	worse	predicted_SO -1.328
SD	Jackson Merrill	batter / CF	worse	predicted_AB -1.3

Structural audit findings

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	COL pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	COL
info	CURRENT_PITCHER_DUPLICATION	SD pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	SD
info	CURRENT_FINALIZED_MATCH_BOX	COL model pitcher projections differ from final simulation pitching box score (ER: current=7, sim=8).	COL
info	CURRENT_FINALIZED_MATCH_BOX	COL model-prediction batting totals differ from final simulation box score (AB: current=34, sim=36; R: current=4, sim=5; BB: current=2, sim=1; SO: current=8, sim=11; HBP: current=0, sim=1).	COL
info	CURRENT_PLAYER_MATCH_BOX	COL current finalized hitter lines are model predictions and differ from the simulation box score: Hunter Goodman: H current=1 box=2, HR current=1 box=0, SO current=1 box=0; Cole Carrigg: H current=1 box=2, 2B current=1 box=0, HR current=0 box=1, RBI current=1 box=2; Connor Norby: AB current=4 box=5, H current=1 box=2, 2B current=0 box=2, R current=0 box=1; Mickey Moniak: RBI current=1 box=0, SO current=1 box=2; Adael Amador: H current=1 box=0, RBI current=1 box=0, R current=1 box=0, HBP current=0 box=1; Jake McCarthy: AB current=4 box=5, 2B current=1 box=0, RBI current=1 box=2, SO current=1 box=2; TJ Rumfield: AB current=4 box=3, H current=1 box=0; Kyle Karros: R current=0 box=1	COL
info	CURRENT_FINALIZED_MATCH_BOX	SD model-prediction batting totals differ from final simulation box score (AB: current=32, sim=38; H: current=9, sim=11; 2B: current=2, sim=0; 3B: current=0, sim=3; HR: current=1, sim=2; RBI: current=5, sim=8; R: current=5, sim=8; BB: current=3, sim=2).	SD
info	CURRENT_PLAYER_MATCH_BOX	SD current finalized hitter lines are model predictions and differ from the simulation box score: Ty France: HR current=1 box=0, RBI current=1 box=0, BB current=0 box=1; Fernando Tatis Jr.: AB current=4 box=5, H current=1 box=3, 2B current=1 box=0, 3B current=0 box=1; Luis Campusano: H current=1 box=0, RBI current=1 box=0, BB current=1 box=0, SO current=1 box=2; Manny Machado: AB current=4 box=5, H current=1 box=2, HR current=0 box=1, RBI current=1 box=3; Dustin Harris: AB current=3 box=5, H current=1 box=2, 3B current=0 box=1, RBI current=0 box=1; Jackson Merrill: 2B current=1 box=0, 3B current=0 box=1, RBI current=1 box=0; Xander Bogaerts: AB current=3 box=4, R current=0 box=1, BB current=1 box=0, SO current=0 box=2; Jake Cronenworth: AB current=3 box=4, HR current=0 box=1, RBI current=0 box=3, R current=0 box=1	SD
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.

Cross-game lever tension

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

Day over day: did it improve or worsen

Compared against packet 20260913. Last recorded change: v4.30.0 (2026-09-08) Hitter shrinkage is derived from the between-hitter spread rather than fitted to slate error, so an established extreme profile is no longer flattened toward the league.

The adjustment the system made in response

Luna deep post-mortem (full)

1. Evidence receipt and gaps

Source receipt

All seven source-manifest entries were reviewed as declared:

Source	Status	Declared purpose
`prediction_review_20260914.json`	Loaded	Official outcomes and scoring errors
`PREDICTION_SYSTEM_DEEP_AUDIT_20260914.json`	Loaded	Probability, provenance, starter, and decision contracts
`continuous_learning_daily_20260914.json`	Loaded	Stage quality, calibration, lineage, score distribution, and starter comparisons
`detailed_20260914.html`	Loaded	Full game and simulation rendering
`20260914_ultra_detailed_player_audit.html`	Loaded	Player-level prediction trace
`audit_report_20260914.html`	Loaded	Invariant findings
`audit_report_20260914_integrity_revision.html`	Missing	Post-Luna remediation proof for immutable revision artifacts

The missing integrity-revision artifact is the only explicit source-manifest gap. It does not invalidate the game outcomes or structural audits, but it prevents verification of the claimed post-remediation immutable-revision proof.

Slate coverage

The slate contained 10 games, and all 10 were reviewed:

ATL@CHC, BAL@NYM, CHW@CLE, DET@TOR, LAD@CIN, MIA@ARI, NYY@MIN, SD@COL, SEA@LAA, and SF@STL.

Coverage was complete for the ordinary prediction artifacts:

Expected games: 10
Canonical artifacts loaded: 10
Continuous-learning games: 10
Forensic deep-audit games: 10
Structural audits completed: 10
Starter comparisons: 10
Games structurally audited: 10

The aggregate structural audit reports 0 critical, 0 error, 0 warning, and 74 informational findings. Every game-level structural audit passed with zero defects.

Internal contradictions and gaps

Missing integrity-revision audit. The manifest marks the artifact missing even though all ordinary structural audits passed.
Probability-publication lineage is incomplete for two games. BAL@NYM and MIA@ARI have unexplained calibrated-to-published changes of at least two percentage points. This is explicitly registered as critical. It is not yet evidence that the governor was bypassed.
The lineage contains smaller unexplained deltas. ATL@CHC has a published-minus-calibrated delta of -1.1176 percentage points, marked nonmaterial and noncritical. The register’s critical threshold is ≥2 percentage points.
Starter comparison is technically complete but not semantically complete for every starter. CHW@CLE lacks a settled projection line for Sean Newcomb; MIA@ARI lacks one for Corbin Burnes. Therefore, “10/10 starter comparisons” means the game had a starter-comparison record, not that every starter had a complete projected line.
The reported “confidence edge” is not uniformly the half-distance from 50%. The game diagnostics explicitly warn against treating a 0.12 edge as a near coin flip. The published probability and the signed home-minus-away edge must be interpreted separately.
Simulation evidence is operationally weak. The simulation generated zero winner flips and zero flip candidates, and this has persisted for 11 slates. That is a persistent diagnostic/code concern, not evidence that simulation had no theoretical value.

---

2. Executive diagnosis

Overall result

Winner accuracy: 7/10, 70%
Misses: BAL@NYM, CHW@CLE, MIA@ARI
Model-validation winner accuracy: 59.82%
Walk-forward accuracy: 55.94%
Slate result versus walk-forward baseline: better by roughly 14 percentage points, but one slate is not sufficient to establish durable improvement.
Mean model-margin absolute error: 2.2 runs
Directional cover rate: 7/10
Weighted absolute calibration error: 0.2812, poor for a 10-game settled sample
Exact run-model holdout R²: extremely weak:
Away runs: 0.0210
Home runs: 0.0118
Total runs: 0.0207

The winner result was acceptable, but it was materially better than the score and probability quality. Correct winners must not be treated as evidence that the score forecasts, confidence levels, or internal causal pipeline were sound.

Winner accuracy

The model correctly selected seven winners:

ATL@CHC
DET@TOR
LAD@CIN
NYY@MIN
SD@COL
SEA@LAA
SF@STL

It missed:

BAL@NYM: strong 0.26 home edge, wrong side
CHW@CLE: 0.01 home edge, toss-up, wrong side
MIA@ARI: 0.03 away edge, toss-up, wrong side

The strongest winner call was NYY@MIN, with a 0.14 away edge, strong negative home composite evidence, and a correct 8–3 prediction direction. However, the model’s top-confidence selection lost on this slate: BAL@NYM was the top-confidence miss. The loss streak for the top-confidence pick is currently 1 slate.

Calibration quality

Calibration was poor on this slate:

0.50–0.55 bucket: average confidence 0.5142, 4/6 hits, observed hit rate 0.6667; reported error -0.1525 under the stated definition.
0.55–0.60 bucket: average confidence 0.5767, 3/3 hits, observed hit rate 1.0; reported error -0.4233.
0.60–0.65 bucket: average confidence 0.6275, 0/1 hits, observed hit rate 0; reported error +0.6275.
Weighted absolute calibration error: 0.2812 across 10 games.

The register correctly classifies this as high, not critical: the evidence shows poor single-slate alignment, but the sample is too small to justify changing calibration weights immediately. The high-raw cluster contained three games and went 2/3, but its empirical hit rates were only approximately 0.536–0.569 despite calibrated probabilities of 0.6705–0.7259. This is a warning against high raw confidence, but not a complete rolling calibration curve.

Score accuracy

Score accuracy was substantially weaker than winner accuracy:

Seven games had predicted margins of two runs or fewer.
Three actual margins were at least four runs.
Mean margin error: 2.2 runs
Total/team-run errors were large in several games:
BAL@NYM: team-run error 8, total bias +8
LAD@CIN: team-run error 6, total bias +6
MIA@ARI: team-run error 6, total bias -6
SF@STL: team-run error 6, total bias +6

The validated run models explain almost none of holdout run variance. This is a high model risk and a persistent 11-slate issue, not a critical evidence-contract failure.

Starter accuracy

Starter projections were mixed:

Three starters exceeded projected ERA by at least 3.00:
Reynaldo López: +9.21 ERA error
Kade Anderson: +8.58
Tomoyuki Sugano: +5.01
Three thin-history starters had fewer than 12 prior appearances in the feed:
Jonah Tong: 4
Kade Anderson: 4
Quinn Mathews: 6

Starter errors were often variance rather than a clear systematic defect. However, SEA@LAA provides a credible process concern: Kade Anderson had only four prior appearances and produced a 10.38 ERA against a 1.80 projection. The registered recommendation to reduce confidence for thin-history starters is directionally reasonable, but the current cross-slate evidence does not establish a global starter-ERA bias. The cross-slate starter-ERA pattern is explicitly marked nonpersistent: only 6 of 12 slates shared the sign, with z = 0.32.

Artifact integrity

Artifact integrity was strong for the loaded artifacts:

All 10 game structural audits passed.
All reported game audits had zero critical/error/warning findings.
Forecast-file integrity was clean in every game.
Learning eligibility was true because all game artifacts passed critical/error gates.

However, integrity is not completely closed because:

The integrity-revision audit artifact is missing.
Two material probability publication deltas are unexplained.
The simulation has been operationally inert for 11 consecutive slates.

Causal issue register reconciliation

Critical: unexplained published probability deltas

ID: unexplained_probability_publication_delta
Severity: Critical
Evidence: Two games have unexplained calibrated-to-published deltas of at least two percentage points.
Affected games: BAL@NYM, MIA@ARI
Action: Block promotion until the missing transformation is recorded and reconciled.
Guardrail: Do not assume the probability governor was bypassed until the pick-side math is traced.

This is a contract/provenance failure, not necessarily a model-quality failure. The data show:

BAL@NYM: calibrated/governed 0.725; published home probability 0.627507.
MIA@ARI: calibrated/governed 0.5676; published home probability 0.517426.

The transformation may be a legitimate confidence-control step, but it is not recorded in the lineage.

High: single-slate calibration instability

ID: single_slate_calibration_instability
Severity: High
Evidence: Weighted absolute calibration error 0.2812 across 10 games.
Affected games: all 10.
Action: Observe the same buckets over the rolling window before changing calibration weights.
Guardrail: Do not infer a stable calibration curve from one slate.

High: weak run-model signal

ID: weak_run_model_signal
Severity: High
Evidence: Home/away/total R² = 0.0118/0.0210/0.0207.
Affected games: all 10.
Action: Treat exact scores as low-confidence; investigate features and target design through isolated training.
Guardrail: Small positive R² does not make exact scores reliable.

Informational: complete audit coverage

ID: complete_slate_audit_coverage
Severity: Info
Evidence: 10/10 games structurally audited.
Action: Preserve per-game audit receipts.
Guardrail: Complete coverage does not cancel findings inside the games.

Informational: simulation inert on slate

ID: simulation_inert_on_slate
Severity: Info at the slate level, but escalated to persistent pattern across slates.
Evidence: Zero flip candidates and zero published flips.
Affected games: all 10.
Action: Measure simulation contribution over rolling slates and determine whether the module is no-oping.
Guardrail: Zero winner flips is not automatically a defect and does not prove zero score-reconciliation work.

---

3. Every game

The signed confidence edge below is the published home-minus-away edge. For away picks, the published pick probability is 1 - published_home_probability.

ATL@CHC — correct winner, wrong score

Actual: ATL 3, CHC 7; winner CHC.
Predicted: ATL 5, CHC 6; winner CHC.
Published pick probability: CHC 0.523824.
Confidence edge: home +0.0476.
Bucket: Lean.
Brier: 0.226744.
Team-run absolute error: 3.
Total-run bias: +1.
Winner status: Correct winner; score wrong.

Evidence edges and controls

Season edge: home +0.1076.
Recent edge: home +0.0113.
Pitcher edge: home +0.0487.
Composite edge: home +0.1692.
Starter-experience control: not applied; both starters had established feed histories.
Recency guardrail: not applied.
Confidence pipeline reached 0.535 before publication controls; published home probability was 0.523824.
Savant adjustment applied.
Score volatility: no change.

Starter comparison

ATL Reynaldo López: projected 3 ER/14 outs, actual 5 ER/9 outs, actual ERA 15.00 versus projected 5.79; ER error +2, ERA error +9.21.
CHC David Peterson: projected 4 ER/13 outs, actual 1 ER/18 outs, actual ERA 1.50 versus projected 8.31; ER error -3, ERA error -6.81.

Score path

Historical ML 4.27–4.61 → variance 4–5 → enhanced 4–7 → Savant 4–7 → fundamental review 4–7 → capped 4–7 → simulation 4–7 → published 5–6.

Player-level adjustments

Pete Crow-Armstrong exceeded projection by +1.39 runs and was identified as an elite-contact profile.
Matt Olson was -1.4 runs versus projection.
Seiya Suzuki was -1.26.
Mauricio Dubón and Michael Conforto contributed positive deviations.

Audit and causes

Structural audit: PASS; 0 defects, 101 good, 7 informational.
Primary cause: no material winner error; score error was driven mainly by starter variance and player deviations.
Secondary cause: López was hit substantially harder than projected; Peterson was much better than projected.
The post-game forensic conclusion called the result a solid correct call. No artifact defect was identified.

---

BAL@NYM — wrong winner and wrong score

Actual: BAL 2, NYM 1; winner BAL.
Predicted: BAL 4, NYM 7; winner NYM.
Published pick probability: NYM 0.627507.
Confidence edge: home +0.255.
Bucket: Strong.
Brier: 0.393765.
Team-run absolute error: 8.
Total-run bias: +8.
Winner status: Wrong winner and wrong score.

Evidence edges and controls

Season edge: home +0.3369.
Recent edge: home +1.000.
Pitcher edge: home -0.0188.
Composite edge: home +0.9339.
Recent NYM form was materially stronger: 7–3 with +22 run differential versus BAL 3–7 and -15.
Starter-experience control was applied because Jonah Tong had only four prior appearances.
Shrink factor: 0.5667.
Confidence before shrink/publication: approximately 0.725.
Published home probability: 0.627507.
Recency guardrail: not applied.
Savant adjustment applied.
Score volatility: supported blowout-margin expansion.

Starter comparison

BAL Brandon Young: projected 3 ER/16 outs, actual 1 ER/15 outs, actual ERA 1.80 versus projected 5.06; ER error -2, ERA error -3.26.
NYM Jonah Tong: projected 1 ER/15 outs, actual 0 ER/18 outs, actual ERA 0.00 versus projected 1.80; ER error -1, ERA error -1.80.

Both starters outperformed the run projections, but the game’s principal failure was the much lower-than-expected overall run environment.

Score path

Historical 4.27–4.61 → variance 4–5 → enhanced 4–7 → Savant approximately 4.03–7.50 → fundamental 4–7 → capped 4–7 → simulation 4–7 → published 4–7.

Player-level adjustments

The system made 24 material player adjustments.
Colton Cowser received a large positive adjustment.
Brett Baty and Francisco Lindor received positive changes.
Pete Alonso, Juan Soto, and Dylan Beavers were materially below projections.
NYM’s lineup produced only one run against a forecast of seven; BAL also produced only two against four.

Audit and causes

Structural audit: PASS; 0 defects, 101 good, 8 informational.
Primary cause: run_environment_over; total projected runs 11 versus actual 3.
Secondary cause: wrong_conviction_side; the system backed NYM with a strong 0.26 edge.
Starter conclusion: variance; Brandon Young’s excellent outing was not explained by a credible venue, opponent, or experience signal.
This is the most consequential single-game miss because the probability was both high and wrong.

---

CHW@CLE — wrong winner and wrong score

Actual: CHW 7, CLE 3; winner CHW.
Predicted: CHW 4, CLE 5; winner CLE.
Published pick probability: CLE 0.505654.
Confidence edge: home +0.0113.
Bucket: Toss-up.
Brier: 0.255686.
Team-run absolute error: 5.
Total-run bias: -1.
Winner status: Wrong winner and wrong score.

Evidence edges and controls

Season edge: home -0.0028.
Recent edge: home +0.302.
Pitcher edge: home +0.0305.
Composite edge: home +0.3211.
Starter-experience control: not applied.
Recency guardrail: applied.
Confidence was approximately 0.508 before later controls; published home probability 0.505654.
Savant adjustment applied.
Score volatility: no change; low-score probability 0.6518.

Starter comparison

CHW Sean Newcomb: actual 0 ER in 3 outs. The settled feed had no projection line; a separate projected ERA of 5.4 appears, but actual_era and formal starter comparison are missing.
CLE Gavin Williams: projected 1 ER/16 outs, actual 2 ER/15 outs, actual ERA 3.60 versus projected 1.69; ER error +1, ERA error +1.91.

Score path

Historical 4.55–4.38 → variance 5–4 → enhanced 4–5, with margin-recency and one-run-margin controls → Savant approximately 3.81–4.98 → fundamental 4–5 → capped 4–5 → simulation 4–5.

Player-level adjustments

Miguel Vargas was +1.4 runs and identified as an elite-contact profile.
Sam Antonacci received a large positive RBI adjustment.
Kyle Teel was +1.4 runs in the forensic accounting.
Munetaka Murakami and Angel Martínez were each -1.4 runs.
The game-level offense favored CHW much more than the projection captured.

Audit and causes

Structural audit: PASS; 0 defects, 102 good, 7 informational.
Primary cause: toss_up_variance.
Secondary cause: lean_side_miss; the edge toward CLE was only 0.0113.
Starter data quality issue: missing settled projection for Newcomb.
This game should not be used to justify a major winner-side change. It was a near-coin-flip outcome, though the documented 0.0113 edge is not the same as a 0.12 moderate edge.

---

DET@TOR — correct winner, wrong score

Actual: DET 6, TOR 5; winner DET.
Predicted: DET 5, TOR 4; winner DET.
Published pick probability: DET 0.494346; home TOR 0.505654.
Confidence edge: home -0.0113, equivalent away +0.0113.
Bucket: Toss-up.
Brier: 0.244378.
Team-run absolute error: 2.
Total-run bias: -2.
Winner status: Correct winner; score wrong.

Evidence edges and controls

Season edge: home -0.4635.
Recent edge: home -0.0353.
Pitcher edge: home -0.0639.
Composite edge: home -0.5843.
Season-quality margin adjustment: -0.1947 toward DET.
Starter-experience control: not applied.
Recency guardrail: applied.
Savant adjustment applied.
Score volatility: no change.

Starter comparison

DET Troy Melton: projected 3 ER/18 outs, actual 2 ER/8 outs, actual ERA 6.75 versus projected 4.50; ER error -1, ERA error +2.25.
TOR José Soriano: projected 5 ER/17 outs, actual 4 ER/19 outs, actual ERA 5.68 versus projected 7.94; ER error -1, ERA error -2.26.

Score path

Historical 4.55–4.38 → variance 5–4 → enhanced 4–5 → Savant 3.81–4.98 → fundamental 4–5 → capped 4–5 → simulation 4–5.

Player-level adjustments

The documented hitter deviations were modest. Dillon Dingler was -0.93 runs, while Spencer Torkelson was +0.78.

Audit and causes

Structural audit: PASS; 0 defects, 100 good, 8 informational.
Primary cause: none assigned.
Secondary cause: none assigned.
Win quality: toss_up_went_our_way.
This is a useful counterexample against aggressive reweighting: starter and team-run estimates were relatively close, and the winner direction was correct.

---

LAD@CIN — correct winner, wrong score

Actual: LAD 4, CIN 1; winner LAD.
Predicted: LAD 8, CIN 3; winner LAD.
Published pick probability: LAD 0.580000; home CIN 0.420000.
Confidence edge: home -0.160, equivalent away +0.160.
Bucket: Moderate.
Brier: 0.1764.
Team-run absolute error: 6.
Total-run bias: +6.
Winner status: Correct winner; wrong score.

A 0.16 away edge is moderate, not near coin-flip territory.

Evidence edges and controls

Season edge: home -1.8942.
Recent edge: home -1.462.
Pitcher edge: home -0.0291.
Composite edge: home -2.3713.
Season-quality margin adjustment: -3.0307 toward LAD.
Starter-experience control: not applied.
Recency guardrail: applied.
Confidence path reached approximately 0.726 before later controls, but the published away probability was 0.58.
Score-volatility action: supported blowout-margin expansion.

Starter comparison

LAD Tarik Skubal: projected 3 ER/20 outs, actual 0 ER/21 outs; ER error -3, ERA error -4.05.
CIN Nick Lodolo: projected 7 ER/13 outs, actual 4 ER/21 outs; ER error -3, ERA error -9.40.

Both starters were better than their projections, especially Lodolo relative to an unusually high projected ERA.

Score path

Historical approximately 5–4 → variance 6–4 → enhanced 8–3 → Savant approximately 7.78–2.83 → fundamental 8–3 → capped 8–3 → simulation 8–3.

Player-level adjustments

Kyle Tucker and Teoscar Hernández each added approximately +1.26 runs.
Max Muncy was -1.73.
Mookie Betts and Will Smith were below projection.
The winner was supported by strong season/recent edges, but the run total and margin were overstated.

Audit and causes

Structural audit: PASS; 0 defects, 100 good, 8 informational.
Primary cause: no winner error; score overprojection.
Secondary cause: starter variance, especially Lodolo’s unexpectedly effective outing.
Win quality: right_winner_wrong_score.

---

MIA@ARI — wrong winner and wrong score

Actual: MIA 7, ARI 8; winner ARI.
Predicted: MIA 5, ARI 4; winner MIA.
Published pick probability: MIA 0.517426; home ARI 0.482574.
Confidence edge: home -0.0349, equivalent away +0.0349.
Bucket: Toss-up.
Brier: 0.267730.
Team-run absolute error: 6.
Total-run bias: -6.
Winner status: Wrong winner and wrong score.

Evidence edges and controls

Season edge: home +0.165.
Recent edge: home +0.078.
Pitcher edge: home +0.0107.
Composite edge: home -0.0462.
Starter-experience control applied because one starter had one prior appearance in the relevant feed.
Shrink factor: 0.4042.
The trace is internally awkward: it reports probability before control at 0.432368 and after control at 0.472665, while another field reports governed probability 0.5676 and published home probability 0.517426. This reinforces the registered lineage defect.
Recency guardrail: not applied.
Score volatility: no change.

Starter comparison

MIA Sandy Alcantara: projected 4 ER/18 outs, actual 4 ER/15 outs; ER error 0, ERA error +1.20.
ARI Corbin Burnes: no settled starter projection line; a projected ERA of 9.0 appears, while the formal actual ERA is absent. The overall starter record reports 3 projected ER/9 outs and actual 5 ER/14 outs, but the per-player feed is incomplete.

Score path

Historical 4.80–4.71 → variance 5–5 → enhanced 5–4 → Savant approximately 5.49–4.19 → fundamental 5–4 → capped 5–4 → simulation 5–4.

Player-level adjustments

ARI hitters were systematically under-projected in the result:

Lars Nootbaar: +1.54 runs, elite contact profile.
Jordan Lawlar: +1.40, above-average profile.
Corbin Carroll: +0.76.
MIA’s Jakob Marsee: +2.16.
Javier Sanoja: +1.71.

This was a high-variance offensive game, with both teams exceeding many individual projections.

Audit and causes

Structural audit: PASS; 0 defects, 101 good, 7 informational.
Primary cause: run_environment_under; actual total 15 versus projected 9.
Secondary cause: toss_up_variance.
Probability-lineage integrity defect is independently registered for this game.
The result does not support a broad offense or starter parameter change without more evidence.

---

NYY@MIN — correct winner, wrong score

Actual: NYY 8, MIN 3; winner NYY.
Predicted: NYY 6, MIN 4; winner NYY.
Published pick probability: NYY 0.570000; home MIN 0.430000.
Confidence edge: home -0.140, equivalent away +0.140.
Bucket: Moderate.
Brier: 0.1849.
Team-run absolute error: 3.
Total-run bias: -1.
Winner status: Correct winner; wrong score.

The 0.14 away edge is moderate and should not be described as a near coin flip.

Evidence edges and controls

Season edge: home -1.2593.
Recent edge: home -0.8513.
Pitcher edge: home +0.0048.
Composite edge: home -1.2549.
Season-quality margin adjustment: -0.5289 toward NYY.
Starter-experience control: not applied.
Recency guardrail: applied.
Score-volatility action: supported blowout-margin expansion.

Starter comparison

NYY Will Warren: projected 3 ER/16 outs, actual 0 ER/17 outs; ER error -3, ERA error -5.06.
MIN Dean Kremer: projected 3 ER/14 outs, actual 1 ER/17 outs; ER error -2, ERA error -4.20.

Both starters outperformed projections.

Score path

Historical approximately 4–5 → variance 5–5 → enhanced 6–4 → Savant approximately 6.20–4.15 → fundamental 6–4 → capped 6–4 → simulation 6–4.

Player-level adjustments

José Caballero and Austin Wells each added approximately +1.4 runs.
Josh Bell added +1.4 for MIN.
Kody Clemens and Ryan Jeffers were below projection.
MIN’s bullpen allowed six runs, a major source of score error not captured by the starter lines.

Audit and causes

Structural audit: PASS; 0 defects, 101 good, 7 informational.
Primary and secondary causal fields were empty because the winner call was strong.
Forensic note identified venue-split evidence for Warren: his away ERA was worse than his other-side ERA. The registered recommendation is to shadow-test context.venue_adjustment_weight.
Win quality: strong_correct_call.

---

SD@COL — correct winner, wrong score

Actual: SD 8, COL 7; winner SD.
Predicted: SD 8, COL 5; winner SD.
Published pick probability: SD 0.580000; home COL 0.420000.
Confidence edge: home -0.160, equivalent away +0.160.
Bucket: Moderate.
Brier: 0.1764.
Team-run absolute error: 2.
Total-run bias: -2.
Winner status: Correct winner; wrong score.

Evidence edges and controls

Season edge: home -1.366.
Recent edge: home -1.246.
Pitcher edge: home -0.0346.
Composite edge: home -1.6789.
Season-quality margin adjustment: -0.5737 toward SD.
Starter-experience control: not applied.
Recency guardrail: applied.
Score-volatility action: supported blowout-margin expansion.

Starter comparison

SD Casey Mize: projected 3 ER/15 outs, actual 2 ER/15 outs; ER error -1, ERA error -1.80.
COL Tomoyuki Sugano: projected 3 ER/14 outs, actual 6 ER/15 outs; ER error +3, ERA error +5.01.

Sugano was the principal pitching miss.

Score path

Historical approximately 5–5 → variance 6–5 → enhanced 8–5 → Savant approximately 7.74–4.67 → fundamental 8–5 → capped 8–5 → simulation 8–5.

Player-level adjustments

Jake McCarthy exceeded projection by +1.09.
Cole Carrigg exceeded projection by +1.09.
Luis Campusano was -0.80.
Hunter Goodman was +0.80 but did not fully account for the final run environment.

Audit and causes

Structural audit: PASS; 0 defects, 100 good, 8 informational.
Primary/secondary causal fields were empty because the winner call was correct.
Forensic attribution classified Sugano’s result as an outlier, with no credible signal explaining the 5.01 ERA error.
Win quality: strong_correct_call.

---

SEA@LAA — correct winner, wrong score

Actual: SEA 4, LAA 6; winner LAA.
Predicted: SEA 5, LAA 6; winner LAA.
Published pick probability: LAA 0.527588.
Confidence edge: home +0.0552.
Bucket: Lean.
Brier: 0.223173.
Team-run absolute error: 1.
Total-run bias: +1.
Winner status: Correct winner; score wrong.

Evidence edges and controls

Season edge: home -0.2584.
Recent edge: home -0.326.
Pitcher edge: home -0.0359.
Composite edge: home -0.1226.
Starter-experience control applied because Kade Anderson had only four prior appearances.
Shrink factor: 0.5667.
Published home probability after control: 0.538184 in the trace; lineage later lists 0.527588.
Recency guardrail: applied.
Score volatility: no change.

Starter comparison

SEA Kade Anderson: projected 1 ER/15 outs, actual 5 ER/13 outs; ER error +4, ERA error +8.58.
LAA Reid Detmers: projected 2 ER/19 outs, actual 3 ER/18 outs; ER error +1, ERA error +1.66.

Score path

Historical approximately 4–4 → variance 5–4 → enhanced 5–6 → Savant approximately 5.23–6.04 → fundamental 5–6 → capped 5–6 → simulation 5–6.

Player-level adjustments

Jose Siri: +1.25 runs.
J.P. Crawford: +1.07.
Christian Moore: -0.80.
LAA lineup exceeded its projection by three runs.

Audit and causes

Structural audit: PASS; 0 defects, 102 good, 7 informational.
Primary/secondary causal fields were empty because the winner was correct.
Starter uncertainty was real and documented. The forensic recommendation was to reduce confidence when prior feed history is thin and validate any run-parameter change in shadow.
Win quality: solid_correct_call.

---

SF@STL — correct winner, wrong score

Actual: SF 1, STL 2; winner STL.
Predicted: SF 4, STL 5; winner STL.
Published pick probability: STL 0.505000.
Confidence edge: home +0.0100.
Bucket: Toss-up.
Brier: 0.245025.
Team-run absolute error: 6.
Total-run bias: +6.
Winner status: Correct winner; wrong score.

Evidence edges and controls

Season edge: home +0.4205.
Recent edge: home -0.238.
Pitcher edge: home +0.0357.
Composite edge: home +0.4303.
Starter-experience control applied to Quinn Mathews, with six prior appearances; shrink factor 0.675.
Recency guardrail: applied.
Score volatility: no change.

Starter comparison

SF Landen Roupp: projected 1 ER/16 outs, actual 2 ER/13 outs; ER error +1, ERA error +2.46.
STL Quinn Mathews: projected 2 ER/15 outs, actual 1 ER/20 outs; ER error -1, ERA error -2.25.

Score path

Historical approximately 4–5 → variance 4–5 → enhanced 4–5 → Savant approximately 3.85–4.99 → fundamental 4–5 → capped 4–5 → simulation 4–5.

Player-level adjustments

Joshua Baez: -1.4 runs.
Shay Whitcomb: -0.93.
Leonardo Bernal: +0.92.
Both lineups under-produced materially.

Audit and causes

Structural audit: PASS; 0 defects, 101 good, 7 informational.
Primary/secondary causal fields were empty.
Win quality: toss_up_went_our_way.
This game is a direct warning against reducing all run errors to starter problems: the primary deviation was low offense on both sides.

---

4. Miss reconstruction

Earliest wrong-direction stage for each loss

BAL@NYM

Historical ML: home advantage was already present.
Variance stage: home remained ahead.
Enhanced prediction: home score expanded to 7 and the model moved toward a stronger NYM conviction.
Savant/environment stage: home remained strongly favored.
Fundamental review, cap, and simulation did not reverse that direction.
Earliest wrong-direction movement: enhanced prediction, where recent form, head-to-head/low-score logic, and related inputs increased the NYM advantage before the game’s actual low-scoring realization.
Failure type: primarily run-environment miss, secondarily bad weighting/conviction escalation, with irreducible starter/offensive variance.
The confidence publication lineage is separately defective and must not be conflated with the winner-direction error.

CHW@CLE

Historical ML was nearly even and slightly favored CHW by runs.
Variance stage moved to CHW 5–4.
Enhanced prediction reversed the direction to CLE 5–4.
Later stages did not reverse it.
Earliest wrong-direction stage: enhanced prediction, where recent form, matchup recency, and one-run-margin controls moved from CHW toward CLE.
Failure type: bad weighting/control interaction in a genuine toss-up, plus starter/input incompleteness because Newcomb lacked a settled projection line, and irreducible variance.
The published probability remained near 50%; this was not a high-conviction failure.

MIA@ARI

Historical ML was nearly even, slightly favoring MIA.
Variance stage became 5–5.
Enhanced prediction moved to MIA 5–4.
Savant moved both teams upward but preserved MIA’s lead.
Later stages preserved the wrong MIA direction.
Earliest wrong-direction stage: enhanced prediction, with the model underweighting ARI’s eventual offensive output and preserving a low-scoring environment.
Failure type: run-environment miss, offense miss, and toss-up variance. Starter evidence was also incomplete for Burnes.
Probability lineage contains an independent material publication defect.

Loss classification

Game	Bad input	Bad weighting	Calibration	Starter error	Offense/bullpen miss	Run-environment miss	Irreducible variance
BAL@NYM	Incomplete explanation of publication transformation	Yes, conviction escalation	Yes, strong 0.26 edge lost	Mostly variance; both starters better	NYM offense -6 versus projection	Primary; 11 projected runs vs 3 actual	Yes
CHW@CLE	Missing settled Newcomb line	Yes, enhanced stage reversed side	Low-confidence toss-up	Newcomb line unavailable; Williams +1 ER	CHW offense +1, CLE offense -3	Secondary, only -1 total bias	Primary toss-up variance
MIA@ARI	Burnes starter line incomplete	Yes, ARI offense underweighted	Toss-up	Alcantara close; Burnes incomplete	ARI offense +4; bullpen contributed	Primary; 9 projected runs vs 15 actual	Yes

A confidence edge of 0.12, where present on other games, is moderate—not a near coin flip. In this slate, the 0.14 and 0.16 edges on NYY@MIN, LAD@CIN, and SD@COL were meaningful moderate edges, whereas CHW@CLE and SF@STL were approximately 0.01 toss-ups.

Pipeline-stage quality across all stages

Stage	Winner accuracy	Winner hits	Team-run MAE	Total-run MAE	Margin MAE	Change versus historical ML
Historical ML	60%	6/10	2.0828	3.3433	1.8811	Baseline
Variance applied	50%	5/10	1.9500	3.3000	1.7000	Margin/team-run improved; winner regressed
Enhanced prediction	70%	7/10	2.1000	3.4000	2.0000	Winner improved; runs/margin worsened
Embedded Savant environment	60%	6/10	2.1500	3.3000	2.3000	Winner and margin regressed
Fundamental edge review	70%	7/10	2.1000	3.4000	2.2000	Winner recovered; margin remained worse
Capped target	70%	7/10	2.1000	3.4000	2.2000	No change
Simulation result	70%	7/10	2.1000	3.4000	2.2000	No change
Published final	70%	7/10	2.1000	3.4000	2.2000	No change

Important transition findings:

historical_ml → variance_applied: winner accuracy fell from 6/10 to 5/10, with four direction changes to ties and SF@STL explicitly regressing.
variance_applied → enhanced_prediction: winner accuracy recovered to 7/10; CHW@CLE regressed, while DET@TOR, NYY@MIN, and SF@STL improved.
enhanced_prediction → embedded_savant_run_environment: ATL@CHC moved CHC to a tie and winner accuracy fell to 6/10.
embedded_savant → fundamental_edge_target_review: ATL@CHC recovered to CHC and winner accuracy returned to 7/10.
Fundamental review, capping, simulation, and publication made no winner-direction changes after that recovery.

The recovery at the fundamental stage does not erase the intermediate Savant-stage regression. The pipeline had a genuine intermediate winner-direction failure even though the final published winner was restored.

The simulation’s zero-flip result is especially important: it did not independently validate or correct any winner on this slate.

---

5. Cross-game counterfactuals

Lever tension 1: widen run distributions

Lever: score.variance_multiplier / score.variance_floor
Target metrics: Brier score and calibration error.
Helps: BAL@NYM, where a 0.26-confidence pick lost and the forecast was severely overconfident in the projected run environment.
Hurts: None listed in the slate-level tension record.
Net evidence: 1 helps, 0 hurts. This supports a small, bounded shadow increase in variance, but not an unconstrained change. The persistent run-model weakness means widening the distribution should be treated as uncertainty management, not as a way to create predictive signal.

Lever tension 2: pitcher form weights and prior-game shrinkage

Lever: player.pitcher.v4 form weights / player.pitcher_prior_games
Target metrics: starter ERA MAE and starter outs MAE.
Helps: ATL@CHC, CHW@CLE, LAD@CIN, NYY@MIN, SD@COL, SEA@LAA.
Hurts or risks hurting: DET@TOR and SF@STL, where starter ER error was only 1; heavier reweighting could disturb already reasonable projections.
Net evidence: 6 helps versus 2 hurts. This supports a bounded shadow experiment, particularly for thin-history starters, but not a broad unconditional reweighting.

Lever tension 3: recent/season run-mean weighting

Lever: recent.* / season.* run-mean weighting
Target metrics: team-run MAE, total-run MAE, accuracy.
Helps: BAL@NYM, CHW@CLE, LAD@CIN, MIA@ARI, SF@STL.
Hurts or risks hurting: DET@TOR, SD@COL, and SEA@LAA, where team-run error was only 2, 2, and 1.
Net evidence: 5 helps versus 3 hurts. This is mixed evidence and does not justify a direct promotion. Any change should be a small shadow experiment with game-level regression monitoring.

Tempting one-game fixes that must not be made

Do not globally increase NYM confidence because of the strong recent-form signal in BAL@NYM; that would worsen calibration risk after the 0.26-confidence miss.
Do not flip toss-up handling based on CHW@CLE; its published edge was only 0.0113 and the outcome was explicitly classified as toss-up variance.
Do not broadly raise ARI offense weights solely because Nootbaar and Lawlar exceeded projections in MIA@ARI; only 21 of 55 material player deviations were supported by the hitter’s own season contact quality.
Do not lower all starter projections because of Kade Anderson or Sugano; several starter projections were reasonably close, and the cross-slate starter bias is nonpersistent.
Do not use the correct winners in LAD@CIN or SD@COL to claim the run model is sound; both had substantial score error despite correct direction.
Do not treat the zero simulation flips as evidence that simulation should be removed. First determine whether the simulation is no-oping because of the documented early-exit/design issue.

---

6. Data and contract defects

Audit finding inventory

Across the structural audit summaries:

Critical: 0
Errors: 0
Warnings: 0
Informational: 74

Every per-game structural audit passed. No structural invariant failure was recorded in the game artifacts.

The broader forensic findings are nevertheless material:

Critical contract/provenance finding

unexplained_probability_publication_delta

Affected: BAL@NYM, MIA@ARI
Material calibrated-to-published changes lack recorded transformations.
Required action: block promotion until lineage is reconstructed.
Do not call this a governor bypass without tracing pick-side math.

High model findings

single_slate_calibration_instability

WAE: 0.2812.
High model/process risk, not a critical contract failure.
Action: continue rolling-window monitoring.

weak_run_model_signal

Holdout R² near zero for all run targets.
High model risk.
Action: feature/target investigation and isolated backtest.

Informational findings

complete_slate_audit_coverage

Coverage was complete and should remain in every report.

simulation_inert_on_slate

Zero flip candidates and zero published flips.
Informational on this slate, but escalated across 11 consecutive slates.

Prediction-model defects versus presentation/artifact bugs

Prediction-model/process defects

Near-zero run-model signal.
Persistent simulation no-op/inertness.
Poor single-slate probability calibration.
Thin-history starter uncertainty.
Incomplete settled starter lines for Newcomb and Burnes.
Potential overreliance on recent/season run means in the enhanced stage.
Player-level projections have weak separation:

H spread ratio 0.386
H correlation 0.213
2B metric worse than flat baseline: MAE 0.264 versus flat MAE 0.231, with 17.4% overprojection.

Presentation/provenance/artifact defects

Missing integrity-revision audit artifact.
Unexplained publication transformations in two probability lineages.
Inconsistent probability fields in MIA@ARI, where the control trace and lineage show different intermediate values.
Starter-comparison records that exist structurally but lack a complete settled projection line for one starter in two games.

The clean forecast artifact does not prove the underlying model was correct. It only establishes that the saved artifact was not corrupted in the ordinary structural sense.

Complete pick-side probability lineage audit

The required chain is:

raw model → calibrated → governor → score coherence/recency → published

Games with fully recorded transformations

CHW@CLE: published probability matches calibrated/governed value; recency guardrail applied.
DET@TOR: same.
LAD@CIN: score coherence followed by matchup-recency guardrail.
NYY@MIN: same.
SD@COL: same.
SEA@LAA: same, though the detailed trace and lineage should still be reconciled numerically.
SF@STL: same.
ATL@CHC: numerical delta is small and marked nonmaterial, but the text says no recorded transformation explains it.

Materially incomplete lineages

BAL@NYM: raw 0.6845 → calibrated 0.725 → governor 0.725 → score coherence 0.725012, but published home probability 0.627507. The 9.75-point drop is not recorded in the lineage. Starter-experience shrinkage appears in the detailed trace, but the lineage does not label it as the complete publication transformation.
MIA@ARI: raw 0.5242 → calibrated 0.5676 → governor 0.5676 → score coherence 0.567632, but published home probability 0.517426. The 5.02-point drop is unexplained in the lineage. The detailed trace additionally reports inconsistent pre/post values around the starter-experience shrink.

No bypass should be asserted until these transformations are traced, but these are genuine contract defects that block promotion.

---

7. Cross-slate persistence

The trend file contains 11 prior/current slates, with 10 prior slates loaded. The current slate is therefore not isolated for persistent findings.

Finding streaks

Finding	Consecutive-slate count	Present now?	Interpretation
`high_raw_probability_cluster`	0	No	One occurrence in window; not persistent
`pipeline_winner_instability`	0	No	Two occurrences in window, but not current
`score_margin_compression`	0	No	One occurrence in window; not persistent
`simulation_inert_on_slate`	11	Yes	Established persistent failure
`single_slate_calibration_instability`	1	Yes	Current and recurring in 5/11 window slates, but only one consecutive slate
`weak_run_model_signal`	11	Yes	Established persistent failure

The two 11-slate findings are escalated findings and must be treated as established failures, not as single-slate noise.

Escalated finding: simulation inertness

Streak: 11 consecutive slates
Incidence: 11/11 window slates
Target lever listed: simulation.allow_score
Target metrics: team-run MAE and total-run MAE
Parameter actionability: explicitly false
Direction of fix: confirm the simulation is running rather than silently no-oping.

The self-assessment identifies a likely root cause: the simulation uses the same draws for target-realization search and win-probability estimation, and exits early after finding a target-realizing play. With approximately two selected draws, the calculated win rate cannot function as an independent probability. The recommended fix is code/diagnostic work in the simulation distribution builder and draw-loop early exit, not a forced nightly parameter move.

Escalated finding: weak run-model signal

Streak: 11 consecutive slates
Incidence: 11/11 window slates
Listed target levers:
training.xgb.n_estimators
training.xgb.max_depth
training.xgb.learning_rate
Target metrics: team-run MAE and total-run MAE
Parameter actionability: explicitly false
Direction of fix: isolated feature/target investigation and retraining backtest.

The registered evidence is strong:

Total R²: 0.020666
Home R²: 0.011791
Away R²: 0.021038

The recommended remedy is not nightly hyperparameter adjustment. The self-assessment proposes adding market no-vig probability where legitimately sourced, park/weather environment, bullpen rest/fatigue, and platoon-adjusted lineup quality, followed by same-fingerprint holdout testing and a model-schema version bump.

Current non-escalated calibration finding

single_slate_calibration_instability:

Present current slate.
Consecutive streak: 1.
Incidence: 5 of 11 window slates.
Target levers: confidence.empirical_max_weight, confidence.base_cap, confidence.moderate_cap.
Target metrics: Brier score and calibration error.

This is concerning, but the causal register correctly says not to change calibration weights from one slate alone. The self-assessment’s auto-action proposal to lower caps is therefore best treated as a bounded shadow experiment, not an unmonitored promotion.

Metric trends

margin_bias: improving, mean -1.4185. The series ends at -0.4.
mean_team_run_error_on_losses: regressing, mean 5.6132. Current value 6.333.
weighted_calibration_error: regressing, mean 0.1579. Current value 0.2812.
winner_accuracy: improving, mean 0.6185. Current value 0.7000.

These trends show an important divergence: winner accuracy and margin bias are improving, while calibration and loss-score error are worsening.

Top-confidence pick loss streak

Current streak: 1
Recent sequence ends: hit on 20260913, miss on 20260914.
This is not yet a persistent top-confidence failure.

Cross-slate bias holds

The two explicit cross-slate bias findings remain nonactionable:

Starter ERA error:

6/12 slates share the sign.
Mean 0.144, z = 0.32.
Status: hold.

Lineup run delta:

7/12 slates share the sign.
Mean -0.163, z = -0.82.
Status: hold.

They should not be treated as established persistent biases.

---

8. Adjustments for today

Recommendations are ranked by persistence, scope, expected metrics, affected games, and regression risk.

1. Fix and independently validate the simulation module

Evidence strength: Highest; 11 consecutive slates.
Registered lever/action: simulation.allow_score is listed, but the finding explicitly says parameter actionability is false.
Change: Do not force a parameter move. Perform the registered code/diagnostic action: separate target-realization search from independent win-distribution estimation, remove or redesign the early exit, and record the batch size.
Expected metric: Improve the validity of simulation contribution; potentially improve team-run MAE and total-run MAE. Winner-flip rate should become measurable rather than structurally zero.
Helps: All games in future slates; on this slate it would most directly provide an independent check for BAL@NYM, CHW@CLE, and MIA@ARI.
Hurts: No direct lever-tension harm is registered, but an unvalidated simulation could introduce noise. Use the proposed shadow gate and rollback monitor.
Regression risk: Medium until independent draws are validated.
Reason not to force a parameter: The registered finding says this is a code/diagnostic failure, not a tunable nightly parameter.

2. Run an isolated feature/target backtest for the run model

Evidence strength: Highest; 11 consecutive slates and near-zero R².
Registered targets: training.xgb.n_estimators, training.xgb.max_depth, training.xgb.learning_rate, but the finding marks parameter actionability false.
Change: No direct hyperparameter promotion. Add and test the registered feature-contract changes: legitimate market no-vig probability, park/weather environment, bullpen rest/fatigue, and platoon-adjusted lineup quality. Compare same-fingerprint holdout R² and MAE; bump model schema if semantics change.
Expected metric: Increase holdout R² and reduce team-run and total-run MAE.
Helps: All games, especially high-run-error games BAL@NYM, LAD@CIN, MIA@ARI, and SF@STL.
Hurts: Could degrade close, already reasonable games such as DET@TOR, SD@COL, and SEA@LAA if promoted without gating.
Regression risk: High if promoted without a holdout and shadow comparison.
Reason not to force a parameter: The evidence says no bounded nightly knob can repair a feature/target problem.

3. Bounded calibration-cap shadow experiment

Evidence strength: Moderate; current WAE 0.2812, current high-conviction miss, and calibration metric is regressing across the window, but the consecutive streak is only 1.
Exact registered parameter: confidence.base_cap.
Change: Reduce confidence.base_cap by a small amount, for example 0.02, in shadow only.
Expected metric: Lower Brier score and weighted calibration error; reduce high-conviction miss severity.
Helps: Most directly BAL@NYM, where the published 0.6275 home probability still overstated the true side despite the starter-experience shrink. It could also help high-raw clusters such as LAD@CIN and SD@COL by reducing overstatement, although those winners were correct.
Hurts: The lever-tension register lists no direct harmed games on this slate. The principal risk is underconfidence on valid moderate/strong calls such as NYY@MIN.
Regression risk: Moderate. Require rolling calibration and Brier monitoring before promotion.
Guardrail: This is a shadow experiment, not a claim that one slate proves a cap change is necessary.

4. Bounded run-variance shadow experiment

Evidence strength: Moderate; the current slate has large score misses and the lever tension shows one help versus zero hurts, while margin bias is improving.
Exact registered parameter: score.variance_multiplier.
Change: Increase by a small amount, for example 0.05, in shadow only.
Expected metric: Reduce score-tail overconfidence and improve Brier/calibration behavior in high-run-error games; potentially improve team-run and total-run MAE through less compressed distributions.
Helps: BAL@NYM most clearly; the lever tension explicitly identifies it as helped by wider distributions.
Hurts: No direct harmed game is listed for this slate, but it could make already close score forecasts less precise in DET@TOR, SD@COL, and SEA@LAA.
Regression risk: Moderate to high because the run model has near-zero signal; widening uncertainty cannot create predictive information.
Guardrail: Shadow only, with rollback if team-run MAE or total-run MAE worsens over the registered window.

5. Thin-history starter shadow control

Evidence strength: Moderate and game-specific; strongest evidence is SEA@LAA, plus thin-history starters in BAL@NYM and SF@STL.
Exact registered parameter: player.pitcher_prior_games.
Change: Increase the effective prior-history requirement modestly for starters with fewer than 12 feed appearances, or equivalently strengthen the prior-games shrinkage in shadow.
Expected metric: Reduce starter ERA MAE and confidence error for thin-history starters.
Helps: SEA@LAA most directly; also future games analogous to BAL@NYM and SF@STL.
Hurts: Could reduce useful responsiveness for genuinely improved young starters, and the lever-tension record warns that DET@TOR and SF@STL had accurate starter projections.
Regression risk: Moderate.
Guardrail: Do not apply globally to established starters; gate by documented feed-history count and validate starter outs as well as ERA.

Defects that require abstention rather than tuning

Do not tune around MIA@ARI until its probability lineage and Burnes starter record are reconciled.
Do not promote any confidence change until the missing publication transformation is recorded for BAL@NYM and MIA@ARI.
Do not use score.variance_floor as a substitute for fixing the near-zero run-model feature signal.
Do not force a parameter for the simulation finding because the registered finding explicitly identifies a code/diagnostic action.

---

9. Bottom line

Known

The slate produced a 7–3 winner record.
Winner accuracy was better than the validated baseline on this slate, but one slate cannot establish durable improvement.
The three losses were:

one strong-conviction run-environment failure: BAL@NYM;
one genuine toss-up with a small wrong-side lean: CHW@CLE;
one toss-up with a six-run underprojection and incomplete starter data: MIA@ARI.

Exact run prediction is weak. Holdout R² is approximately 0.01–0.02, and this weakness has persisted for 11 slates.
Simulation contribution is not independently trustworthy and has been inert for 11 slates.
Calibration is poor on the current slate and is trending worse across the monitoring window.
Ordinary artifact coverage is complete and all structural audits passed.
Two probability lineages have critical unexplained publication transformations.
The integrity-revision audit artifact is missing.

Uncertain

Whether the unexplained probability deltas represent a legitimate but undocumented starter/recency transformation or an actual publication-contract bypass.
Whether recent/season run weighting is beneficial after controlling for slate composition; the current lever tension is mixed.
Whether starter prior weighting should be globally changed; the cross-slate starter bias is not persistent.
Whether widening variance will improve realized score metrics or merely make forecasts less specific; it should remain a shadow experiment.
The extent to which the reported player-level deviations are learnable: only 21 of 55 material deviations were supported by the hitter’s own season contact quality.

Safest single next action

Do not promote any model or probability change. First reconcile the complete pick-side probability lineage for BAL@NYM and MIA@ARI, while independently validating the simulation distribution code in shadow.

That is the safest immediate action because it addresses the only critical contract failure and the strongest persistent operational failure before allowing calibration or variance adjustments to mask a provenance or no-op simulation defect.

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	CIN pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	CIN
info	CURRENT_PITCHER_DUPLICATION	LAD pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	LAD
info	CURRENT_FINALIZED_MATCH_BOX	CIN model pitcher projections differ from final simulation pitching box score (ER: current=7, sim=8).	CIN
info	CURRENT_FINALIZED_MATCH_BOX	CIN model-prediction batting totals differ from final simulation box score (H: current=7, sim=5; 2B: current=1, sim=3; RBI: current=4, sim=3; R: current=4, sim=3; BB: current=3, sim=2; SO: current=11, sim=8).	CIN
info	CURRENT_PLAYER_MATCH_BOX	CIN current finalized hitter lines are model predictions and differ from the simulation box score: Tyler Stephenson: H current=1 box=0, RBI current=1 box=0, BB current=1 box=0, SO current=1 box=2; Dane Myers: H current=1 box=0, 2B current=1 box=0, BB current=1 box=0; Sal Stewart: H current=1 box=0, RBI current=1 box=0, R current=1 box=0; Juan Brito: H current=1 box=2, 2B current=0 box=1, HR current=0 box=1, RBI current=0 box=3; Elly De La Cruz: H current=1 box=0, HR current=1 box=0, RBI current=1 box=0, R current=1 box=0; TJ Friedl: SO current=1 box=0; Ke'Bryan Hayes: AB current=3 box=1, R current=0 box=1, BB current=0 box=2, SO current=1 box=0; JJ Bleday: AB current=3 box=4, H current=1 box=0, SO current=1 box=2	CIN
info	CURRENT_FINALIZED_MATCH_BOX	LAD model-prediction batting totals differ from final simulation box score (AB: current=32, sim=39; H: current=8, sim=12; RBI: current=4, sim=8; R: current=5, sim=8; BB: current=4, sim=5).	LAD
info	CURRENT_PLAYER_MATCH_BOX	LAD current finalized hitter lines are model predictions and differ from the simulation box score: Tommy Edman: H current=1 box=3, HR current=0 box=1, RBI current=1 box=4, BB current=0 box=1; Mookie Betts: RBI current=1 box=0, SO current=0 box=1; Kyle Tucker: AB current=4 box=5, BB current=1 box=0, SO current=1 box=0; Will Smith: AB current=4 box=5, RBI current=0 box=1, R current=0 box=1, BB current=1 box=0; Enrique Hernández: AB current=3 box=5, 2B current=0 box=1, RBI current=0 box=2, R current=0 box=1; Josue De Paula: H current=0 box=1, BB current=0 box=1, SO current=1 box=0; Teoscar Hernández: AB current=4 box=5, H current=1 box=2, RBI current=1 box=0, R current=0 box=1; Miguel Rojas: AB current=3 box=4, H current=1 box=0, 2B current=1 box=0, BB current=0 box=1	LAD
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.

Team	Starter	Projected	Actual	ER err
MIA	Sandy Alcantara	4 ER / 18 outs	4 ER / 15 outs	0
ARI	Corbin Burnes	3 ER / 9 outs	5 ER / 14 outs	2

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
MIA	Sandy Alcantara (Starter)	15 / 18	7.2 / 6	4 / 4	7 / 7	1 / 1	2 / 2	1.81	6 / 5	3.16	venue away: 4.32 vs 3.14 ERA (worse here) · opp run factor 0.99 (average) · 31 prior appearances this season
MIA	Calvin Faucher (Closer)	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0.53	0 / 2	0.95
MIA	Michael Petersen (Middle Reliever (High Leverage))	3 / 3	0 / 0	0 / 0	2 / 0	0 / 0	0 / 0	0.45	3 / 0	1.1
MIA	Tyler Zuber (Setup Man)	3 / 3	18 / 0 (+18.00)	2 / 0	1 / 0	1 / 0	1 / 0	0.42	1 / 0	1.08

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
Jakob Marsee (CF)	MIA	4 / 4	3 / 1	0 / 0	1 / 0	1 / 0	1 / 0	2 / 1	0 / 1	0 / 1	2.16	0.308 average	no signal
Javier Sanoja (3B)	MIA	4 / 4	2 / 1	0 / 1	1 / 0	1 / 0	3 / 1	2 / 1	0 / 0	0 / 0	1.71	0.292 below average	good day
Lars Nootbaar (LF)	ARI	5 / 4	3 / 1	0 / 0	0 / 0	1 / 0	2 / 0	2 / 0	0 / 1	1 / 1	1.54	0.349 elite	projection too low
Jordan Lawlar (CF)	ARI	4 / 3	2 / 1	0 / 0	0 / 0	1 / 0	2 / 0	2 / 0	0 / 0	1 / 1	1.4	0.328 above average	projection too low
Corbin Carroll (RF)	ARI	5 / 4	2 / 1	0 / 1	0 / 0	1 / 0	3 / 0	1 / 1	0 / 1	1 / 1	0.76	0.341 elite	projection too low
Ketel Marte (2B)	ARI	5 / 4	1 / 1	1 / 0	0 / 0	0 / 1	0 / 1	0 / 1	0 / 0	0 / 0	-0.62	0.346 elite	bad day
Heriberto Hernández (LF)	MIA	4 / 4	1 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	1 / 0	1 / 1	-0.6	0.352 elite	bad day
Otto Lopez (SS)	MIA	5 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	3 / 0	-0.47	0.327 above average	—
Agustín Ramírez (DH)	MIA	4 / 3	1 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	2 / 1	0.47	0.285 below average	—
Griffin Conine (RF)	MIA	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 1	1 / 1	0 / 1	3 / 1	-0.33	0.368 elite	—
Joe Mack (C)	MIA	2 / 3	1 / 1	0 / 0	0 / 0	0 / 0	1 / 0	1 / 0	1 / 0	0 / 1	0.33	0.289 below average	—
Xavier Edwards (2B)	MIA	5 / 4	2 / 1	0 / 0	0 / 0	0 / 0	1 / 0	0 / 0	0 / 1	2 / 1	0.14	0.313 average	—
Geraldo Perdomo (SS)	ARI	2 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	2 / 1	2 / 0	-0.14	0.326 above average	—
Tim Tawa (1B)	ARI	3 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	1 / 0	1 / 0	0 / 1	-0.14	0.314 average	—
Gabriel Moreno (C)	ARI	3 / 4	1 / 1	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	1 / 0	1 / 0	0.02	0.359 elite	—
Kyle Stowers (1B)	MIA	5 / 4	1 / 1	1 / 1	0 / 0	0 / 0	1 / 1	1 / 0	0 / 0	1 / 1	0	0.328 above average	—
Nolan Arenado (3B)	ARI	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 1	0 / 0	3 / 1	0	0.293 below average	—

Signal	Home-away edge	Meaning
Season quality	0.165	positive favors home; negative favors away
Recent form	0.078	positive favors home; negative favors away
Starting-pitcher matchup	0.011	positive favors home; negative favors away
Composite	-0.046	positive favors home; negative favors away

Stage	Away-home runs	Confidence	What changed
historical_ml	4.8-4.71	n/a	Base ML historical model prediction
variance_applied	5-5	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	5-4	0.517	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by starter_quality_favours_the_other_side, projected_margin_is_one_run)
embedded_savant_run_environment	5-4	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	5-4	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	5-4	n/a	Capped at max 12 realistic runs
simulation_result	5-4	n/a	Final inning-by-inning simulation result

Stage	Before	After	Toward final pick	What changed
variance_applied	4.8-4.71	5-5	-0.092	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	5-5	5-4	1	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by starter_quality_favours_the_other_side, projected_margin_is_one_run)
embedded_savant_run_environment	5-4	5-4	-0	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	5-4	5-4	-0	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	5-4	5-4	-0	Capped at max 12 realistic runs
simulation_result	5-4	5-4	-0	Final inning-by-inning simulation result

Team	Player	Role	Net effect	Largest change
ARI	Tim Tawa	batter / 1B	worse	predicted_SO 2.119
MIA	Heriberto Hernández	batter / LF	worse	predicted_H 2.09
MIA	Agustín Ramírez	batter / DH	worse	predicted_H 1.7
MIA	Otto Lopez	batter / SS	worse	predicted_H -1.64
MIA	Kyle Stowers	batter / 1B	worse	predicted_RBI 1.51
ARI	Corbin Carroll	batter / RF	better	predicted_RBI 1.5
MIA	Xavier Edwards	batter / 2B	worse	predicted_SO 1.434
ARI	Gabriel Moreno	batter / C	worse	predicted_H -1.4

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	ARI pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	ARI
info	CURRENT_PITCHER_DUPLICATION	MIA pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	MIA
info	CURRENT_FINALIZED_MATCH_BOX	ARI model-prediction batting totals differ from final simulation box score (AB: current=33, sim=34; H: current=8, sim=7; BB: current=3, sim=2; SO: current=5, sim=7).	ARI
info	CURRENT_PLAYER_MATCH_BOX	ARI current finalized hitter lines are model predictions and differ from the simulation box score: Lars Nootbaar: H current=1 box=0, BB current=1 box=0; Geraldo Perdomo: H current=1 box=2, RBI current=0 box=1, R current=0 box=1, BB current=1 box=0; Corbin Carroll: AB current=4 box=3, 2B current=1 box=0, HR current=0 box=1, RBI current=0 box=2; Gabriel Moreno: H current=1 box=0, 2B current=1 box=0, RBI current=1 box=0, R current=1 box=0; Pavin Smith: H current=0 box=1, R current=0 box=1, BB current=0 box=1, SO current=0 box=1; Ketel Marte: 2B current=0 box=1, HR current=1 box=0, RBI current=1 box=0, R current=1 box=0; Nolan Arenado: H current=1 box=2, 2B current=0 box=1, SO current=1 box=0; Tim Tawa: AB current=3 box=4, H current=1 box=0, RBI current=1 box=0, SO current=1 box=3	ARI
info	CURRENT_FINALIZED_MATCH_BOX	MIA model-prediction batting totals differ from final simulation box score (AB: current=34, sim=35; 2B: current=2, sim=1; HR: current=1, sim=3; RBI: current=4, sim=5; BB: current=3, sim=2; SO: current=7, sim=10).	MIA
info	CURRENT_PLAYER_MATCH_BOX	MIA current finalized hitter lines are model predictions and differ from the simulation box score: Kyle Stowers: AB current=4 box=5, 2B current=1 box=0, HR current=0 box=1, RBI current=1 box=2; Xavier Edwards: H current=1 box=0, BB current=1 box=0, SO current=1 box=2; Heriberto Hernández: H current=1 box=3, 2B current=0 box=1, HR current=1 box=0, RBI current=1 box=0; Griffin Conine: BB current=1 box=0, SO current=1 box=2; Otto Lopez: H current=1 box=0, R current=1 box=0, SO current=0 box=1; Jakob Marsee: AB current=4 box=3, H current=1 box=0, R current=1 box=0; Javier Sanoja: H current=1 box=0, 2B current=1 box=0, RBI current=1 box=0, R current=1 box=0; Joe Mack: AB current=3 box=4, HR current=0 box=1, RBI current=0 box=1, R current=0 box=1	MIA
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.

Team	Starter	Projected	Actual	ER err
NYY	Will Warren	3 ER / 16 outs	0 ER / 17 outs	-3
MIN	Dean Kremer	3 ER / 14 outs	1 ER / 17 outs	-2

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
NYY	Will Warren (Starter)	17 / 16	0 / 5.06 (-5.06)	0 / 3	5 / 2	0 / 2	2 / 2	2.15	6 / 4	6.05	venue away: 4.45 vs 3.86 ERA (worse here) · opp run factor 1.04 (average) · 29 prior appearances this season
NYY	Tim Hill (Middle Reliever (High Leverage))	3 / 2	9 / 13.5 (-4.50)	1 / 1	1 / 1	1 / 0	0 / 1	0.26	0 / 1	0.48
MIN	Dean Kremer (Starter)	17 / 14	1.59 / 5.79 (-4.20)	1 / 3	2 / 6	1 / 3	1 / 1	2.08	5 / 5	4.24	venue home: 5.2 vs 6.11 ERA (ok) · opp run factor 1.03 (strong) · 14 prior appearances this season
MIN	A.J. Minter (Middle Reliever (High Leverage))	0 / 4	n/a / 20.25	3 / 3	3 / 4	1 / 0	0 / 1	0.17	0 / 1	0.88
MIN	Andrew Morris (Middle Reliever (High Leverage))	1 / 3	81 / 0 (+81.00)	3 / 0	3 / 0	0 / 0	0 / 0	0.36	1 / 0	0.83
MIN	Jeff Hoffman (Setup Man)	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0.37	1 / 1	1.48

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
José Caballero (2B)	NYY	4 / 3	2 / 1	0 / 0	0 / 0	1 / 0	2 / 0	2 / 0	0 / 0	0 / 1	1.4	0.28 weak	good day
Austin Wells (C)	NYY	4 / 3	1 / 0	0 / 0	0 / 0	1 / 0	1 / 0	1 / 0	0 / 0	2 / 1	1.4	0.288 below average	good day
Josh Bell (DH)	MIN	4 / 4	2 / 1	0 / 0	0 / 0	1 / 0	1 / 1	2 / 0	0 / 0	1 / 1	1.4	0.333 above average	projection too low
Kody Clemens (2B)	MIN	5 / 4	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	1 / 1	0 / 0	1 / 1	-1.4	0.337 above average	bad day
Ryan Jeffers (C)	MIN	4 / 4	0 / 1	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	1 / 1	-1.11	0.357 elite	bad day
Luis García Jr. (1B)	NYY	4 / 4	0 / 1	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 0	1 / 0	-0.78	0.353 elite	bad day
Spencer Jones (CF)	NYY	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	2 / 1	-0.47	0.331 above average	—
Cody Bellinger (LF)	NYY	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 1	1 / 1	0 / 1	0 / 1	-0.33	0.338 above average	—
Aaron Judge (RF)	NYY	4 / 4	1 / 1	0 / 0	0 / 0	1 / 1	3 / 1	1 / 1	0 / 1	2 / 1	-0.33	0.403 elite	—
George Lombard Jr. (SS)	NYY	4 / 3	1 / 1	1 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0 / 0	2 / 1	0.31	0.265 weak	—
Emmanuel Rodriguez (LF)	MIN	4 / 4	1 / 1	0 / 0	1 / 0	0 / 0	2 / 0	0 / 1	0 / 1	2 / 1	0.29	0.607 elite	—
Luke Keaschall (RF)	MIN	2 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	3 / 1	0 / 0	0.19	0.315 average	—
Ben Rice (DH)	NYY	3 / 4	2 / 1	0 / 1	0 / 0	0 / 0	2 / 1	1 / 1	1 / 1	0 / 1	0.16	0.375 elite	—
Brooks Lee (3B)	MIN	4 / 4	1 / 1	0 / 1	0 / 0	0 / 0	0 / 1	0 / 0	1 / 0	0 / 1	0.02	0.275 weak	—
Ryan McMahon (3B)	NYY	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0	0.301 below average	—
Victor Caratini (1B)	MIN	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0	0.318 average	—
Walker Jenkins (CF)	MIN	4 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0	0.255 weak	—
Ryan Kreidler (SS)	MIN	4 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0	0.289 below average	—

Signal	Home-away edge	Meaning
Season quality	-1.259	positive favors home; negative favors away
Recent form	-0.851	positive favors home; negative favors away
Starting-pitcher matchup	0.005	positive favors home; negative favors away
Composite	-1.255	positive favors home; negative favors away

Stage	Away-home runs	Confidence	What changed
historical_ml	4.19-4.35	n/a	Base ML historical model prediction
variance_applied	4-4	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	6-4	0.57	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by matchup_recency_guardrail)
embedded_savant_run_environment	6-4	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	6-4	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	6-4	n/a	Capped at max 12 realistic runs
simulation_result	6-4	n/a	Final inning-by-inning simulation result

Team	Player	Role	Net effect	Largest change
MIN	Ryan Jeffers	batter / C	worse	predicted_RBI 2.25
NYY	Austin Wells	batter / C	worse	predicted_H 1.57
NYY	Ryan McMahon	batter / 3B	boosted	predicted_H 1.44
NYY	Luis García Jr.	batter / 1B	better	predicted_RBI 1.42
MIN	Kody Clemens	batter / 2B	worse	predicted_AB -1.35
NYY	Aaron Judge	batter / RF	better	predicted_BB 1.328
MIN	Ryan Kreidler	batter / SS	worse	predicted_SO 1.325
MIN	Walker Jenkins	batter / CF	better	predicted_AB -1.154

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	MIN pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	MIN
info	CURRENT_PITCHER_DUPLICATION	NYY pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	NYY
info	CURRENT_FINALIZED_MATCH_BOX	MIN model-prediction batting totals differ from final simulation box score (AB: current=33, sim=30; H: current=8, sim=3; 2B: current=2, sim=1; HR: current=1, sim=2; SO: current=7, sim=6; HBP: current=0, sim=1).	MIN
info	CURRENT_PLAYER_MATCH_BOX	MIN current finalized hitter lines are model predictions and differ from the simulation box score: Luke Keaschall: AB current=4 box=3, H current=1 box=0, R current=1 box=0, BB current=1 box=0; Ryan Jeffers: H current=1 box=2, HR current=0 box=1, RBI current=1 box=3, BB current=1 box=0; Emmanuel Rodriguez: AB current=4 box=3, H current=1 box=0; Victor Caratini: AB current=3 box=4, H current=1 box=0, SO current=0 box=1; Brooks Lee: H current=1 box=0, 2B current=1 box=0, RBI current=1 box=0, SO current=1 box=0; Josh Bell: H current=1 box=0, RBI current=1 box=0; Kody Clemens: AB current=4 box=3, H current=1 box=0, HR current=1 box=0, RBI current=1 box=0; Walker Jenkins: AB current=3 box=2, H current=0 box=1, HR current=0 box=1, RBI current=0 box=1	MIN
info	CURRENT_FINALIZED_MATCH_BOX	NYY model-prediction batting totals differ from final simulation box score (AB: current=32, sim=37; H: current=8, sim=10; HR: current=1, sim=3; RBI: current=4, sim=6; R: current=5, sim=6; BB: current=3, sim=2).	NYY
info	CURRENT_PLAYER_MATCH_BOX	NYY current finalized hitter lines are model predictions and differ from the simulation box score: Ben Rice: AB current=4 box=5, H current=1 box=0, 2B current=1 box=0, RBI current=1 box=0; Cody Bellinger: AB current=4 box=5, H current=1 box=2, RBI current=1 box=0, BB current=1 box=0; Aaron Judge: AB current=4 box=3, BB current=1 box=2, SO current=1 box=0; Luis García Jr.: RBI current=1 box=2, SO current=0 box=1; Spencer Jones: HR current=0 box=1, RBI current=0 box=1, R current=0 box=1; George Lombard Jr.: AB current=3 box=4, H current=1 box=0, R current=1 box=0, SO current=1 box=2; José Caballero: AB current=3 box=4, HR current=0 box=1, RBI current=0 box=1, R current=0 box=1; Austin Wells: AB current=3 box=4, H current=0 box=2	NYY
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.

Team	Starter	Projected	Actual	ER err
SEA	Kade Anderson	1 ER / 15 outs	5 ER / 13 outs	4
LAA	Reid Detmers	2 ER / 19 outs	3 ER / 18 outs	1

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
SEA	Kade Anderson (Starter)	13 / 15	10.38 / 1.8 (+8.58)	5 / 1	7 / 5	0 / 1	0 / 0	1.99	5 / 5	3.81	venue away: 4.15 vs 4.32 ERA (ok) · opp run factor 1.01 (weak) · thin sample: 4 prior appearances this season (high uncertainty)
LAA	Reid Detmers (Starter)	18 / 19	4.5 / 2.84	3 / 2	7 / 4	1 / 0	1 / 2	0.92	8 / 7	8.03	venue home: 3.21 vs 3.47 ERA (ok) · opp run factor 1.05 (weak) · 29 prior appearances this season
LAA	Ben Joyce (Closer)	3 / 3	0 / 0	0 / 0	0 / 1	0 / 0	2 / 0	0.66	0 / 1	1.28
LAA	Ryan Watson (Setup Man)	4 / 3	0 / 9 (-9.00)	0 / 1	0 / 1	0 / 1	0 / 1	0.23	2 / 2	1

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
Jose Siri (CF)	LAA	4 / 4	3 / 1	1 / 0	0 / 0	0 / 0	2 / 1	2 / 0	0 / 0	0 / 1	1.25	0.305 average	no signal
J.P. Crawford (SS)	SEA	5 / 4	2 / 1	0 / 0	0 / 0	1 / 0	1 / 0	1 / 0	0 / 1	1 / 1	1.07	0.322 above average	projection too low
Christian Moore (2B)	LAA	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 1	3 / 1	-0.8	0.245 weak	projection too high
Michael Arroyo (DH)	SEA	3 / 3	1 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	1 / 1	0.47	0.447 elite	—
Josh Naylor (1B)	SEA	4 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	-0.47	0.316 average	—
Cole Young (2B)	SEA	4 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	1 / 1	-0.47	0.317 average	—
Weston Wilson (3B)	SEA	3 / 3	1 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	2 / 1	0.47	0.261 weak	—
Randy Arozarena (LF)	SEA	4 / 4	2 / 1	1 / 0	0 / 0	0 / 0	0 / 1	1 / 1	0 / 1	0 / 1	0.45	0.347 elite	—
Julio Rodríguez (CF)	SEA	3 / 4	0 / 1	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	1 / 0	2 / 1	-0.45	0.335 above average	—
Cal Raleigh (C)	SEA	4 / 4	1 / 1	0 / 0	0 / 0	1 / 1	3 / 1	1 / 1	0 / 1	2 / 2	-0.33	0.312 average	—
Zach Neto (SS)	LAA	4 / 4	2 / 1	0 / 0	1 / 0	0 / 1	1 / 1	2 / 1	0 / 1	1 / 2	-0.17	0.31 average	—
Vaughn Grissom (1B)	LAA	2 / 4	1 / 1	0 / 1	0 / 0	0 / 0	1 / 1	0 / 0	1 / 0	0 / 1	0.02	0.305 average	—
Dominic Canzone (RF)	SEA	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 1	1 / 0	0 / 0	2 / 1	0	0.352 elite	—
Mike Trout (DH)	LAA	3 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	1 / 1	1 / 1	0	0.377 elite	—
Kyren Paris (LF)	LAA	2 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0	0.225 weak	—
Wade Meckler (RF)	LAA	3 / 3	1 / 1	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0 / 0	0 / 1	0	0.314 average	—
Travis d'Arnaud (C)	LAA	4 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	0	0.244 weak	—
Denzer Guzman (3B)	LAA	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	2 / 0	0 / 0	0 / 0	0 / 1	0	0.267 weak	—

Signal	Home-away edge	Meaning
Season quality	-0.258	positive favors home; negative favors away
Recent form	-0.326	positive favors home; negative favors away
Starting-pitcher matchup	-0.036	positive favors home; negative favors away
Composite	-0.123	positive favors home; negative favors away

Deep slate report — 20260914

Process, end to end

Slate summary

What drove the losses

Source coverage receipt

High-level causal issue register

Cross-slate trends (persistent-pattern memory)

Escalated persistent patterns

Finding streaks across the window

Metric direction across the window

Pipeline stage regression

Probability lineage and calibration

Score distribution and margin compression

Per-game evidence coverage

Every game, in depth (10)

ATL@CHCHIT

Player-level breakdown (projected vs actual)

Fundamental evidence entering the pick

Score and confidence evolution

Official postgame stage attribution

Largest player-level adjustments from the ultra-detailed trace

Structural audit findings

BAL@NYMMISS

Player-level breakdown (projected vs actual)

Fundamental evidence entering the pick

Score and confidence evolution

Official postgame stage attribution

Largest player-level adjustments from the ultra-detailed trace

Structural audit findings

CHW@CLEMISS

Player-level breakdown (projected vs actual)

Fundamental evidence entering the pick

Score and confidence evolution

Official postgame stage attribution

Largest player-level adjustments from the ultra-detailed trace

Structural audit findings

DET@TORHIT

Player-level breakdown (projected vs actual)

Fundamental evidence entering the pick

Score and confidence evolution

Official postgame stage attribution

Largest player-level adjustments from the ultra-detailed trace

Structural audit findings

LAD@CINHIT

Player-level breakdown (projected vs actual)

Fundamental evidence entering the pick

Score and confidence evolution

Official postgame stage attribution

Largest player-level adjustments from the ultra-detailed trace

Structural audit findings

MIA@ARIMISS

Player-level breakdown (projected vs actual)

Fundamental evidence entering the pick

Score and confidence evolution

Official postgame stage attribution

Largest player-level adjustments from the ultra-detailed trace

Structural audit findings

NYY@MINHIT

Player-level breakdown (projected vs actual)

Fundamental evidence entering the pick

Score and confidence evolution

Official postgame stage attribution

Largest player-level adjustments from the ultra-detailed trace

Structural audit findings

SD@COLHIT

Player-level breakdown (projected vs actual)

Fundamental evidence entering the pick

Score and confidence evolution

Official postgame stage attribution

Largest player-level adjustments from the ultra-detailed trace

Structural audit findings

SEA@LAAHIT

Player-level breakdown (projected vs actual)

Fundamental evidence entering the pick

Score and confidence evolution

Official postgame stage attribution

Largest player-level adjustments from the ultra-detailed trace

Structural audit findings

SF@STLHIT

Player-level breakdown (projected vs actual)

Stage	Away-home runs	Confidence	What changed
historical_ml	3.45-5.4	n/a	Base ML historical model prediction
variance_applied	3-5	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	5-6	0.528	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by projected_margin_is_one_run, matchup_recency_guardrail)
embedded_savant_run_environment	5-6	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	5-6	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	5-6	n/a	Capped at max 12 realistic runs
simulation_result	5-6	n/a	Final inning-by-inning simulation result

Stage	Before	After	Toward final pick	What changed
variance_applied	3.45-5.4	3-5	0.054	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	3-5	5-6	-1	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by projected_margin_is_one_run, matchup_recency_guardrail)
embedded_savant_run_environment	5-6	5-6	0	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	5-6	5-6	0	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	5-6	5-6	0	Capped at max 12 realistic runs
simulation_result	5-6	5-6	0	Final inning-by-inning simulation result

Team	Player	Role	Net effect	Largest change
SEA	Cal Raleigh	batter / C	worse	predicted_SO 2.693
LAA	Kyren Paris	batter / LF	better	predicted_H 2
LAA	Zach Neto	batter / SS	worse	predicted_AB -1.96
SEA	Julio Rodríguez	batter / CF	better	predicted_H 1.66
SEA	Cole Young	batter / 2B	worse	predicted_H 1.66
SEA	Josh Naylor	batter / 1B	worse	predicted_AB -1.53
SEA	Michael Arroyo	batter / DH	worse	predicted_AB 1.5
SEA	J.P. Crawford	batter / SS	better	predicted_RBI 1.5

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	LAA pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	LAA
info	CURRENT_PITCHER_DUPLICATION	SEA pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	SEA
info	CURRENT_FINALIZED_MATCH_BOX	LAA model-prediction batting totals differ from final simulation box score (AB: current=33, sim=32; H: current=7, sim=8; 2B: current=1, sim=0; RBI: current=4, sim=6; R: current=3, sim=6; BB: current=3, sim=2; SO: current=10, sim=8; HBP: current=0, sim=2).	LAA
info	CURRENT_PLAYER_MATCH_BOX	LAA current finalized hitter lines are model predictions and differ from the simulation box score: Mike Trout: BB current=1 box=0; Jose Siri: AB current=4 box=3, H current=1 box=0, RBI current=1 box=0, SO current=1 box=2; Vaughn Grissom: 2B current=1 box=0, RBI current=1 box=2, R current=0 box=1, SO current=1 box=0; Christian Moore: AB current=4 box=3, H current=1 box=0, RBI current=1 box=0, SO current=1 box=0; Kyren Paris: AB current=3 box=4, H current=0 box=2, RBI current=0 box=2, R current=0 box=1; Wade Meckler: AB current=3 box=4, R current=1 box=0; Travis d'Arnaud: AB current=3 box=4, H current=0 box=1, HR current=0 box=1, RBI current=0 box=1; Denzer Guzman: AB current=4 box=3, H current=1 box=0, RBI current=0 box=1, R current=0 box=1	LAA
info	CURRENT_FINALIZED_MATCH_BOX	SEA model-prediction batting totals differ from final simulation box score (AB: current=33, sim=36; H: current=7, sim=9; 2B: current=1, sim=3; RBI: current=4, sim=5; BB: current=3, sim=4; SO: current=10, sim=11; HBP: current=0, sim=1).	SEA
info	CURRENT_PLAYER_MATCH_BOX	SEA current finalized hitter lines are model predictions and differ from the simulation box score: Dominic Canzone: AB current=4 box=3, 2B current=0 box=1, R current=0 box=1, BB current=0 box=1; Cal Raleigh: AB current=4 box=5, H current=1 box=0, HR current=1 box=0, RBI current=1 box=0; Randy Arozarena: AB current=4 box=5, H current=1 box=0, RBI current=1 box=0, R current=1 box=0; Michael Arroyo: AB current=3 box=4, R current=1 box=0; Julio Rodríguez: AB current=4 box=5, H current=1 box=3, HR current=0 box=1, RBI current=1 box=2; J.P. Crawford: H current=1 box=2, 2B current=0 box=1, RBI current=0 box=2, R current=0 box=1; Josh Naylor: AB current=4 box=3, H current=1 box=0, BB current=0 box=1, SO current=1 box=0; Cole Young: AB current=3 box=4, H current=1 box=2	SEA
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.

Team	Starter	Projected	Actual	ER err
SF	Landen Roupp	1 ER / 16 outs	2 ER / 13 outs	1
STL	Quinn Mathews	2 ER / 15 outs	1 ER / 20 outs	-1

Pitcher	Team	Outs	ERA	ER	H	HR	BB	exp BB	SO	exp SO	Context (as-of)
SF	Landen Roupp (Starter)	13 / 16	4.15 / 1.69	2 / 1	5 / 4	0 / 1	3 / 3	2.7	4 / 3	4.39	venue away: 4.28 vs 3.69 ERA (worse here) · opp run factor 1 (average) · 28 prior appearances this season
SF	Trent Harris (Setup Man)	3 / 4	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0.35	1 / 1	1.4
SF	Carson Seymour (Middle Reliever (High Leverage))	3 / 0	0 / n/a	0 / 1	1 / 3	0 / 0	0 / 0	0.67	0 / 2	0.89
SF	Joel Kuhnel (Reliever)	5 / 0	0 / n/a	0 / 2	0 / 2	0 / 0	0 / 1	n/a	1 / 1	n/a
STL	Quinn Mathews (Starter)	20 / 15	1.35 / 3.6	1 / 2	4 / 5	0 / 0	2 / 0	2.01	8 / 4	4.93	venue home: 1.59 vs 4.68 ERA (ok) · opp run factor 0.96 (weak) · 6 prior appearances this season
STL	Justin Bruihl (Middle Reliever (High Leverage))	1 / 3	0 / 0	0 / 0	0 / 1	0 / 0	0 / 0	0.31	0 / 0	0.85
STL	Riley O'Brien (Closer)	3 / 7	0 / 0	0 / 0	1 / 0	0 / 0	1 / 1	0.5	2 / 1	1.07
STL	Ryne Stanek (Middle Reliever (High Leverage))	3 / 2	0 / 0	0 / 0	1 / 2	0 / 0	1 / 0	0.39	2 / 0	1.09

Batter	Team	AB	H	2B	3B	HR	RBI	R	BB	SO	Run impact	Season xwOBA	Signal
Joshua Baez (LF)	STL	4 / 4	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	0 / 0	1 / 1	-1.4	0.316 average	no signal
Shay Whitcomb (2B)	SF	3 / 4	1 / 1	0 / 0	0 / 0	0 / 1	0 / 1	0 / 1	0 / 0	1 / 1	-0.93	0.27 weak	projection too high
Leonardo Bernal (C)	STL	4 / 4	3 / 1	1 / 0	0 / 0	0 / 0	1 / 1	0 / 1	0 / 1	0 / 1	0.92	0.372 elite	projection too low
Scott Bandura (RF)	SF	2 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	2 / 1	-0.47	0.209 weak	—
Christian Koss (SS)	SF	4 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	2 / 1	-0.47	0.255 weak	—
Thomas Saggese (2B)	STL	4 / 4	2 / 1	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	0 / 0	0 / 1	0.47	0.314 average	—
Iván Herrera (DH)	STL	3 / 4	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	1 / 1	1 / 1	0 / 1	-0.47	0.331 above average	—
Nathan Church (CF)	STL	3 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	-0.47	0.289 below average	—
Turner Hill (DH)	SF	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	0 / 1	0 / 0	0 / 1	1 / 1	-0.33	0.316 average	—
Jonah Cox (CF)	SF	4 / 4	1 / 1	0 / 0	0 / 0	0 / 0	1 / 1	0 / 1	0 / 1	2 / 1	-0.33	0.259 weak	—
Bryce Eldridge (1B)	SF	3 / 4	1 / 1	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	1 / 1	1 / 1	-0.31	0.354 elite	—
Grant McCray (LF)	SF	4 / 3	1 / 1	1 / 0	0 / 0	0 / 0	0 / 0	1 / 0	0 / 0	2 / 1	0.31	0.31 average	—
Alec Burleson (1B)	STL	3 / 4	2 / 1	0 / 1	0 / 0	0 / 0	1 / 1	0 / 1	1 / 1	1 / 1	0.16	0.375 elite	—
Andrew Knizner (C)	SF	2 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	0 / 1	-0.14	0.295 below average	—
Masyn Winn (SS)	STL	2 / 3	0 / 1	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	1 / 0	0 / 1	-0.14	0.298 below average	—
Jordan Walker (RF)	STL	3 / 4	1 / 1	0 / 1	0 / 0	0 / 0	0 / 1	0 / 1	1 / 0	1 / 1	0.02	0.343 elite	—
Brett Harris (3B)	SF	2 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0	0.311 average	—
Nolan Gorman (3B)	STL	3 / 3	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	0 / 0	2 / 1	0	0.279 weak	—

Stage	Away-home runs	Confidence	What changed
historical_ml	4.39-4.42	n/a	Base ML historical model prediction
variance_applied	4-4	n/a	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	4-5	0.505	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by winner_starter_is_below_average, projected_margin_is_one_run, matchup_recency_guardrail)
embedded_savant_run_environment	4-5	n/a	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	4-5	n/a	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	4-5	n/a	Capped at max 12 realistic runs
simulation_result	4-5	n/a	Final inning-by-inning simulation result

Stage	Before	After	Toward final pick	What changed
variance_applied	4.39-4.42	4-4	-0.03	Deterministic baseline target (variance factor 1.0)
enhanced_prediction	4-4	4-5	1	Enhanced ML with recent form, H2H, low-score analysis (confidence shrunk by winner_starter_is_below_average, projected_margin_is_one_run, matchup_recency_guardrail)
embedded_savant_run_environment	4-5	4-5	0	Target score adjusted by temporal Savant lineup quality versus opposing starter quality
fundamental_edge_target_review	4-5	4-5	0	Refreshed score target and win probability after Savant/pitcher/team-strength adjustments
capped_target	4-5	4-5	0	Capped at max 12 realistic runs
simulation_result	4-5	4-5	0	Final inning-by-inning simulation result

Severity	Rule	Finding	Subject
info	CURRENT_PITCHER_DUPLICATION	SF pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	SF
info	CURRENT_PITCHER_DUPLICATION	STL pitcher list appears in both predicted_performance.pitchers and player_predictions_by_position.pitchers; IDs match and were de-duplicated in this audit.	STL
info	CURRENT_FINALIZED_MATCH_BOX	SF model-prediction batting totals differ from final simulation box score (AB: current=31, sim=40; H: current=8, sim=10; 2B: current=1, sim=3; HR: current=1, sim=0; R: current=5, sim=4; BB: current=3, sim=2; SO: current=8, sim=5).	SF
info	CURRENT_PLAYER_MATCH_BOX	SF current finalized hitter lines are model predictions and differ from the simulation box score: Bryce Eldridge: H current=1 box=0, 2B current=1 box=0, RBI current=1 box=0, R current=1 box=0; Shay Whitcomb: AB current=4 box=5, HR current=1 box=0, SO current=1 box=0; Turner Hill: H current=1 box=2, RBI current=1 box=2, R current=0 box=1, SO current=1 box=0; Scott Bandura: AB current=3 box=5, 2B current=0 box=1, R current=1 box=0; Grant McCray: AB current=3 box=4, H current=1 box=2, RBI current=0 box=1, R current=0 box=1; Jonah Cox: AB current=4 box=5, 2B current=0 box=1, RBI current=1 box=0, BB current=1 box=0; Christian Koss: AB current=3 box=4, 2B current=0 box=1, R current=1 box=0; Andrew Knizner: AB current=3 box=4, SO current=1 box=0	SF
info	CURRENT_FINALIZED_MATCH_BOX	STL model-prediction batting totals differ from final simulation box score (AB: current=33, sim=41; H: current=8, sim=11; 2B: current=2, sim=3; HR: current=1, sim=2; RBI: current=4, sim=5; BB: current=3, sim=5).	STL
info	CURRENT_PLAYER_MATCH_BOX	STL current finalized hitter lines are model predictions and differ from the simulation box score: Thomas Saggese: AB current=4 box=5, R current=0 box=1, BB current=0 box=1; Leonardo Bernal: AB current=4 box=5, RBI current=1 box=0, R current=1 box=0, BB current=1 box=0; Joshua Baez: AB current=4 box=5, HR current=1 box=0, RBI current=1 box=0, R current=1 box=0; Alec Burleson: H current=1 box=0, 2B current=1 box=0, RBI current=1 box=0, R current=1 box=0; Jordan Walker: AB current=4 box=5, RBI current=1 box=0, R current=1 box=0, SO current=1 box=0; Iván Herrera: AB current=4 box=5, H current=1 box=2, 2B current=0 box=1, BB current=1 box=0; Nolan Gorman: H current=0 box=2, 2B current=0 box=1, RBI current=0 box=1, R current=0 box=1; Nathan Church: AB current=3 box=4, HR current=0 box=1, RBI current=0 box=1, R current=0 box=1	STL
info	RECONCILIATION_PROPAGATION	No score_target_reconciliation step was used for this game.