Baseball · Track record
Model comparisons
Compare returns and forecast quality over time, using frozen pregame inputs and odds. Coverage stays visible alongside performance.
Models on the selected games
1 unit per bet · 2.0 pp minimum edge · kalshi · same pregame checkpoint
2026-01-01 through 2026-10-02 · 6 published game dates · Revision 7
Published data: 2026-09-19 through 2026-09-26. Dates outside this coverage are not included in these results.
Early sample: the fewest settled forecasts for a model in this view is 54 across 6 published dates. These returns describe observed games; they do not establish a long-term model ranking.
| Model | W–L | Net units | ROI | Brier ↓ | Log loss ↓ | Graded forecasts | Pending bets |
|---|---|---|---|---|---|---|---|
| Tree | 15–17 | +1.24 | 3.9% | 0.2306 | 0.6539 | 54 | 0 (0.00 units) |
| Equal blend | 12–16 | -0.86 | -3.1% | 0.2269 | 0.6461 | 54 | 0 (0.00 units) |
| Tabular | 15–15 | +2.20 | 7.3% | 0.2240 | 0.6396 | 54 | 0 (0.00 units) |
W–L and returns cover qualifying, graded bets. ROI is net units divided by graded stake. Brier and log loss cover all graded forecasts, including games where the model passed; lower is better. A dash means no graded observations.
Compare policies
Staking
Minimum probability edge
Game cohort
Switches use published results for the same models and checkpoint. Your dates stay selected. Unpublished combinations are unavailable.
Cumulative net units
By published month within the selected range. Missing dates are not estimated; returns exclude pending and void bets.
- Tree
- Equal blend
- Tabular
Monthly returns table
| Month | Tree | Equal blend | Tabular |
|---|---|---|---|
| 2026-09 | +1.24 / +1.24 | -0.86 / -0.86 | +2.20 / +2.20 |
Coverage and unresolved games
54 of 74 scheduled games have eligible comparison evidence. 20 games were excluded before model comparison because evidence was unavailable.
Why games were excluded
- No quote at the checkpoint: 2
- Checkpoint quote evidence was unavailable: 1
- Quote identity was unavailable: 2
- Shared model inputs were unavailable: 1
- No completed forecast before the checkpoint: 1
- Model inputs were too old: 17
A game can have more than one missing-evidence reason. These games contribute no wins, losses, units or forecast scores.
Model availability below is measured across eligible cases, before same-game cohort exclusions.
| Model | Scored / eligible | Included | Excluded by cohort | Not recorded | Missing inputs | Failed | Passed | Void bets |
|---|---|---|---|---|---|---|---|---|
| Tree | 54 / 54 | 54 | 0 | 0 | 0 | 0 | 22 | 0 (0.00 units) |
| Equal blend | 54 / 54 | 54 | 0 | 0 | 0 | 0 | 26 | 0 (0.00 units) |
| Tabular | 54 / 54 | 54 | 0 | 0 | 0 | 0 | 24 | 0 (0.00 units) |
Confirmed-lineup models
Recorded live forecasts · kalshi · 60 minutes before start
1 unit per bet · Minimum probability edge: 2.0 percentage points
Same-game cohort: only games scored by every selected model enter the performance comparison.