Transparent model evaluation · summary
Model Lab
Compare model descriptions using frozen walk-forward tests and the same random control. Historical evaluation does not prove future performance or change draw odds.
Champion · not promoted
DrawPick Baseline
Frozen version and evaluation report are required before any Smart Pick use.
Challenger
Frequency Blend
Candidate model · no approved evaluation fixture loaded.
Random control
Balanced Random
Uniform control baseline; included in every comparison.
Model standings
| Model | Role | Eligible | Evidence |
|---|---|---|---|
| DrawPick Baseline | Champion candidate | Pending | Frozen version required |
| Frequency Blend | Challenger | Pending | Walk-forward report required |
| Balanced Random | Control | Required | Always included |
Performance over time
No evaluation series is rendered until a reviewed fixture, time split, metric definition, and random-control output exist.
Open advanced comparisonHow evaluation works
- Freeze training window
- Generate candidates
- Settle against held-out draws
- Compare with same random control
- Aggregate across folds
Promotion governance
No automatic promotion. Threshold, minimum sample, stability, methodology review, and owner approval are all required.
Current threshold · not approvedLimitations
Lottery draws are independent random events. Historical patterns may not persist. Evaluation is not proof of predictive skill.