Prime ML + Totals Backtest Dashboard

Synthetic +5% offered-odds backtests, matched model-vs-book checks, and total-goals distribution diagnostics.

Moneyline Read

Full Fav ROI

5.88%

Full Dog ROI

3.27%

Model Logloss

0.6868

Book Logloss

0.7133

Moneyline +5% Cumulative PnL

Moneyline +5% ROI by Season

Moneyline +5% ROI by Odds Bucket

Moneyline Model vs Book Calibration

Totals Read

Line Model Over ROI

5.87%

Line Model Under ROI

6.01%

Line Model Logloss

0.6822

Current Season Over ROI

7.31%

Totals +5% Cumulative PnL

Totals +5% ROI by Season

Totals +5% ROI by Odds Bucket

Totals Main-Line Calibration

Totals Over Probability by Line

Line Read

This page shows only the current promoted totals model. On the full main-line sample, 5.5 is running at 0.07 pp and 6.5 at 0.16 pp. On the current season slice, 5.5 is the bigger sore spot at 0.04 pp, while 6.5 sits at -0.65 pp.

That is why the line-family calibration matters so much: 5.5 and 6.5 do not behave like the same market. The counts are shown under each bar now, because 4.5 and 7.5 main-line samples are tiny and should not be read the same way.

Totals Distribution Comparison

Distribution Note

The raw scalar totals forecast is still under-dispersed. The calibrated totals-goals mean has predicted standard deviation 0.31 versus actual 2.31, which is why the model behaves much better on unders than overs.

The promoted totals model is the line-family + goalie version currently running in prime. The matched book sample is still small, so this page prioritizes the larger synthetic backtest view.