Continuous postgame learning

daily_20261002 · dates 20261002 · generated 2026-10-03T09:00:00.329216+00:00

0-0

Winner picks · —

Comparable starters · 0 identity mismatches

0/0

Starter ER within ±1

0/0

Starter workload within ±1 inning

Slate coverage

Every calendar day in the requested window is listed. Pending means the prediction exists but the official final result has not settled; Missing review artifact identifies a data gap instead of hiding the day.

Date	Games	Settled	Pending	Hits	Misses	Calibration traces	Review status	Pipeline versions
20261002	0	0	0	0	0	0	Available

Prioritized optimization roadmap

Runtime evidence for the six recommendations. “Observed games” counts artifacts generated with the relevant trace; older artifacts remain honestly blank.

Priority	Change	Status	Evidence
P1	Empirical Confidence Blend	Instrumented Waiting For Verified History	applied games: 0formula: w_emp=min(0.5,sample_size/200); final=(1-w_emp)raw+w_empempirical_hit_rate
P1	H2H Base Feature	Implemented And Monitored	available games: 0
P2	Score Distribution Widening	Implemented Above 1.4X	—
P2	Simulation Flip Control	Diagnostic Only Without Independent Support	simulation flip candidates: 0published flips: 0
P3	Recent Form Quality Blend	Configured 10 Not Observed	configured window: 10last 10 games: 0
P3	Continuous Calibration Audit	Waiting For Settled Results	settled games: 0metric: bucket_calibration_error

Confidence calibration by bucket

bucket_calibration_error = average published confidence − observed hit rate. Positive means overconfident; negative means underconfident. Weighted absolute error: .

Bucket	Games	Average confidence	Observed hit rate	Calibration error	Direction

Where the score pipeline gained or lost quality

Direct comparison against official final runs. Positive deltas mean the stage was worse than historical_ml; this is score error, not a synthetic one-game ERA.

Stage	Games	Team-run MAE	Total-run MAE	Margin MAE	Winner accuracy	Δ team MAE vs base	Δ margin MAE vs base

Largest measured regression

null

Automatic questions

Which recorded score stage reinforced losing picks?

insufficient_sample sample=0

No settled miss predictions were available.

Complete structured answer

{
  "status": "insufficient_sample",
  "sample_size": 0,
  "top_stage": null,
  "reason": "No settled miss predictions were available.",
  "ranking": [],
  "interpretation": "No attribution claim can be made without settled games."
}

Which recorded score stage reinforced winning picks?

insufficient_sample sample=0

No settled hit predictions were available.

Complete structured answer

{
  "status": "insufficient_sample",
  "sample_size": 0,
  "top_stage": null,
  "reason": "No settled hit predictions were available.",
  "ranking": [],
  "interpretation": "No attribution claim can be made without settled games."
}

Are Overs performing worse than Unders?

not_measurable sample=0

The current artifacts contain predicted totals but no total line, Over/Under selection, price, or settled total bet.

Complete structured answer

{
  "status": "not_measurable",
  "sample_size": 0,
  "reason": "The current artifacts contain predicted totals but no total line, Over/Under selection, price, or settled total bet.",
  "missing_fields": [
    "total_line",
    "total_selection",
    "odds",
    "settled_result"
  ]
}

Are hitter props outperforming pitcher props?

not_measurable sample=0

Player stat projections are not stored as priced, selected, and settled prop predictions.

Complete structured answer

{
  "status": "not_measurable",
  "sample_size": 0,
  "reason": "Player stat projections are not stored as priced, selected, and settled prop predictions.",
  "missing_fields": [
    "prop_family",
    "prop_line",
    "selection",
    "odds",
    "settled_result"
  ]
}

Which sportsbooks produce better opportunities?

not_measurable sample=0

No settled prediction was linked to a sportsbook snapshot.

Complete structured answer

{
  "status": "not_measurable",
  "sample_size": 0,
  "reason": "No settled prediction was linked to a sportsbook snapshot.",
  "missing_fields": [
    "sportsbook",
    "selection_odds",
    "settled_profit"
  ],
  "books": []
}

Which confidence ranges underperform?

insufficient_sample sample=0

Range	Sample Size	Hits	Misses	Accuracy
[0.000, 0.525)	0	0	0
[0.525, 0.550)	0	0	0
[0.550, 0.600)	0	0	0
[0.600, 1.010)	0	0	0

Complete structured answer

{
  "status": "insufficient_sample",
  "sample_size": 0,
  "minimum_recommended_sample": 20,
  "worst_observed_range": null,
  "ranges": [
    {
      "range": "[0.000, 0.525)",
      "sample_size": 0,
      "hits": 0,
      "misses": 0,
      "accuracy": null
    },
    {
      "range": "[0.525, 0.550)",
      "sample_size": 0,
      "hits": 0,
      "misses": 0,
      "accuracy": null
    },
    {
      "range": "[0.550, 0.600)",
      "sample_size": 0,
      "hits": 0,
      "misses": 0,
      "accuracy": null
    },
    {
      "range": "[0.600, 1.010)",
      "sample_size": 0,
      "hits": 0,
      "misses": 0,
      "accuracy": null
    }
  ]
}

Which monitored features are drifting?

insufficient_sample sample=0

Three slates are descriptive only and cannot establish statistical feature drift.

Complete structured answer

{
  "status": "insufficient_sample",
  "sample_size": 0,
  "minimum_baseline_sample": 30,
  "minimum_recent_sample": 20,
  "reason": "Three slates are descriptive only and cannot establish statistical feature drift.",
  "daily_monitors": [],
  "features_monitored": [
    "confidence",
    "predicted_total",
    "winner_accuracy"
  ]
}

Which model version introduced errors?

insufficient_sample sample=0

Version error claims require at least 30 settled predictions per version.

Complete structured answer

{
  "status": "insufficient_sample",
  "sample_size": 0,
  "minimum_sample_per_version": 30,
  "versions": [],
  "reason": "Version error claims require at least 30 settled predictions per version."
}

Game processing context

How each game moved from the historical model through confidence calibration, H2H and recent-form signals, score distribution and simulation policy. Postgame learning is added only after the official result settles.

Every reviewed game

Date	Game	Pick	Final confidence	Raw confidence	Empirical hit rate	Empirical weight	Pred A-H	Actual A-H	Result	Largest recorded movement	Pipeline	Roadmap signals