Detailed Model Audit

NHL Model Results

This page preserves the expanded NHL model audit for William: the combined live-test card, model-lane split, CLV diagnostics, rejected totals context, and the production blockers that must be solved before a full customer release.

Customer Summary How It Was Tested Back To NHL Hub

Combined Live-Test Candidate

NHL Moneyline + Enriched Team Totals

This is the website-ready lane from the NHL handoff. It combines the full-game moneyline customer-guard lane with the enriched home-team team-total under lane. It is ready for structured 2026-27 live tracking, but not presented as a completed paid-card proof yet.

Combined Profit

+$10,809

+108.09 units at $100 per unit

621 bets | +108.09u | 17.4% ROI

Combined Record

340-281

621 qualified bets | +108.09u

Combined ROI

17.4%

$62,100 risked at $100 per play

CLV

75.9%

Positive CLV | avg decimal CLV +0.0624

Lane	Season	Bets	Record	Profit	Units	ROI	Positive CLV	Avg Decimal CLV
Moneyline + Enriched Team Totals	2024	331	181-150	+$5,457	+54.57u	16.5%	73.4%	+0.0623
Moneyline + Enriched Team Totals	2025	290	159-131	+$5,352	+53.52u	18.5%	78.6%	+0.0625
Combined Live-Test Card	2024-2025	621	340-281	+$10,809	+108.09u	17.4%	75.9%	+0.0624

Lane Split

Moneyline Vs Enriched Team Totals

Model	Bet Type	Bets	Record	Win %	Profit	Units	ROI	Positive CLV	Avg Decimal CLV
Moneyline	Full-game moneyline customer guard	367	188-179	51.2%	+$6,800	+68.00u	18.5%	77.7%	+0.0759
Enriched Team Totals	Home-team team-total unders	254	152-102	59.8%	+$4,008	+40.08u	15.8%	73.2%	+0.0428

Model Notes

What The Audit Preserves

Moneyline Method

L2 logistic regression, strict chronological walk-forward, home-minus-away feature mode, MoneyPuck team stats, goalie starter-proxy features, and historical h2h odds.

Team-Total Method

Enriched market-residual logistic model for home-team unders, adding pregame offense, opponent defense, goalie form/workload, special teams, rest, congestion, and matchup context.

Rejected Totals

Do not include full-game NHL totals on the customer page. They were tested and rejected for now because the forward transfer was not strong enough.

Release Blockers

Need confirmed/probable goalie status at bet time, timestamped live tracking, market movement, slippage, book-specific availability, and data licensing review.