NRL Model Calibration

Heavy favourites ($1.05–1.40), blind

−8.2% ROI

76.0% win rate · n=107

Big underdogs ($3.01+), blind

−21.6% ROI

17.6% win rate · n=108

Win % by market odds

Actual win rate for each price band across all 683 predictions.

Model calibration chart

Each dot = one predicted-probability band. On the dashed line, predicted = actual (perfect calibration). Above the line = model under-confident; below = over-confident.

Trust zone: ~45% to ~60% predicted win probability

NRL is a two-team market — every game has one side near 40–60% implied. The model is best-calibrated in the near-even-money range (45–60% predicted), where predicted and actual win rates sit within ±2pp. Outside that window it over-estimates favourites (says 72%, they win 64%) and slightly under-estimates underdogs in the 35–45% band.

The $1.66–2.40 bracket is the only zone where flat-staking approaches breakeven (−2.2% and +1.0%). Short favourites at $1.05–1.40 lose 8.2% blind because the vig eats the already-thin edge. Underdogs $3.01+ lose 21.6%. The disciplined angle is near-even-money games where the model diverges from the market by ≥3 pts of line, sized with fractional Kelly and graded on CLV.

Win rate & flat ROI by market price band (H2H)

Price band	Games	Winners	Win %	Implied %	Flat ROI
$1.05–1.40	107	81	75.7%	80.8%	−6.4%
$1.41–1.65	124	82	66.1%	68.5%	−3.5%
$1.66–2.00 ★	138	78	56.5%	57.8%	−2.2%
$2.01–2.40 ★	119	54	45.4%	44.9%	+1.0%
$2.41–3.00	87	27	31.0%	37.1%	−16.3%
$3.01–5.00	89	18	20.2%	26.5%	−20.4%
$5.01+	19	2	10.5%	14.6%	−25.8%

★ Only the $1.66–2.40 even-money zone approaches breakeven — −2.2% and +1.0% flat ROI respectively. Every other band loses money blindly backed.

The takeaway: target games where the model rates a team at 50–60% but the market prices them at $1.90–2.20. Size with fractional Kelly, never exceed 2 units on a single H2H position, and judge every bet on closing-line value.

NRL Historical Win Rates 683 predictions