# Conviction Backbone Backtest

Model: `conviction_backbone_v1`
Training calibration before: `2022-07-01`
Backtest from: `2022-07-01`
Matches tested: `21277`

This backtest uses market-implied probabilities as the prior, historical odds-bucket calibration as the correction, and the same longshot/away-win guards used by the live scorer. News is not included in this historical test.

## Avg Price

| Segment | Bets | Strike | Avg odds | Avg edge | P/L | ROI |
|---|---:|---:|---:|---:|---:|---:|
| overall | 0 | n/a | n/a | n/a | 0.000u | n/a |

## Best Price

| Segment | Bets | Strike | Avg odds | Avg edge | P/L | ROI |
|---|---:|---:|---:|---:|---:|---:|
| overall | 67 | 49.3% | 2.635 | 4.7% | 13.440u | 20.1% |
| over_under_2_5 | 61 | 49.2% | 2.578 | 4.4% | 12.100u | 19.8% |
| over_under_2_5:OVER | 30 | 53.3% | 2.634 | 4.1% | 9.140u | 30.5% |
| over_under_2_5:UNDER | 31 | 45.2% | 2.525 | 4.7% | 2.960u | 9.5% |
| 1x2 | 6 | 50.0% | 3.213 | 8.3% | 1.340u | 22.3% |
| 1x2:HOME_WIN | 5 | 60.0% | 2.496 | 8.2% | 2.340u | 46.8% |
| 1x2:DRAW | 1 | 0.0% | 6.800 | 8.9% | -1.000u | -100.0% |

## Conviction Rule

- The backbone does not chase one-off narrative swings.
- Market prior is the anchor. Historical calibration can adjust it, but only within capped movement.
- Team news is an overlay, not a model replacement.
- Long 1X2 prices and away wins need stricter proof because favourite-longshot bias is visible in the local history.
- Over/under 2.5 is the preferred automatic market until Asian handicap history is available.