Simulation Model
ResearchRuns thousands of projected game outcomes using scoring expectations, starters, bullpen context, and frozen sportsbook odds. Selects the strongest edge per game per market.
Research model only. Not official picks. Not part of the public W-L record.
Slate
10/19/2026
Simulations
50,000
Games
1
Research Picks
0
Generated Oct 19, 6:01 PM
Today's Simulation Research Picks
Calibration note: model picks in the 60%–70% confidence range have historically landed at 51% in our sample of 1675 graded picks. Treat as research input, not standalone signal. See the reliability table below for full per-bucket detail.
No simulation research picks available yet. Picks generate after the daily pipeline runs.
Selected-Pick Performance
Record
2325-2283
ROI
-3.5%
Profit
-163.33u
Graded
4608
| Market | W | L | P | ROI |
|---|---|---|---|---|
| Runline | 151 | 150 | 0 | -9.7% |
| Team Total | 837 | 869 | 0 | -5.8% |
| Game Total | 741 | 698 | 0 | +0.6% |
| F5 Runline | 18 | 11 | 0 | +7.7% |
| F5 Total | 578 | 555 | 0 | -3.9% |
Model Variants
The public-facing research model uses a Poisson-based simulation engine. Several challenger variants run in shadow alongside it to test whether different modeling assumptions improve calibration:
Shadow variants are compared internally against the baseline. None are promoted until calibration evidence shows improvement.
Calibration Tracking
Brier Score
0.2699
Log Loss
0.7409
ECE
0.1345
Calibration measures how well the model's predicted probabilities match actual outcomes. A perfectly calibrated model would have events predicted at 70% happening exactly 70% of the time. Lower Brier scores, log loss, and ECE indicate better calibration. This is a research model and is not used for official picks.
| Bucket | Count | Avg Predicted | Actual Rate | Gap |
|---|---|---|---|---|
| 0.00-0.10 | 0 | -- | -- | -- |
| 0.10-0.20 | 0 | -- | -- | -- |
| 0.20-0.30 | 4 | 27.4% | 0.0% | 27.4% |
| 0.30-0.40 | 55 | 36.5% | 36.4% | 0.2% |
| 0.40-0.50 | 161 | 46.5% | 42.9% | 3.6% |
| 0.50-0.60 | 1497 | 55.6% | 47.4% | 8.2% |
| 0.60-0.70 | 1675 | 64.7% | 51.3% | 13.3% |
| 0.70-0.80 | 927 | 74.2% | 54.2% | 20.0% |
| 0.80-0.90 | 247 | 83.5% | 54.3% | 29.3% |
| 0.90-1.00 | 42 | 94.5% | 73.8% | 20.7% |
By Market
Runline
301 graded Brier: 0.2672
Team Total
1706 graded Brier: 0.2763
Game Total
1439 graded Brier: 0.2706
F5 Runline
29 graded Brier: 0.2371
F5 Total
1133 graded Brier: 0.2610
Methodology
The model simulates tens of thousands of possible score outcomes for each game on the daily slate. Each simulation uses expected run totals derived from pitcher matchups, team offensive profiles, and park factors.
Simulated outcomes are aggregated into probability estimates for all seven market types: moneyline, runline, game total, F5 moneyline, F5 runline, F5 total, and team totals.
Those probabilities are compared against frozen sportsbook odds to calculate edge. The model selects the single best-edge pick per game per market type as a research pick.
Selected research picks are tracked in an append-only ledger and graded against actual outcomes. This is separate from the candidate-level audit used for calibration analysis.