Method & What We Found

The rule: when the away team played last night and the home team has had exactly two days off, back the home team. Flat stake, every occurrence, no discretion, no filtering after the fact.

How it is graded: every game is priced at the average US moneyline available when our morning capture job actually runs. Not the closing line, which subscribers cannot bet, and never an assumed −110 — and not a claimed fixed time either: that job's real run time varies by hours day to day, so each game's exact capture time is logged and shown per game in the results below rather than stated as a single round number. This is why our published ROI is lower than most: assumed pricing inflates a 62% win rate by several points.

What the four seasons show: the two-day bucket was profitable in every season tested, including three we had never examined when the rule was first written. Combined: 509 games, 62.1% win rate, +5.6% ROI at real prices, with a 95% confidence interval of −2.2% to +13.3% on that ROI — the figure that actually answers whether a bettor made money. (A narrower-looking comparison exists — win rate's own 95% floor, 57.9%, clears the breakeven implied by the pooled average price, 56.5% — but that comparison assumes every bet was placed at one average price, when it wasn't: season averages alone ranged from −100 to −186. The ROI interval accounts for that price variation and the win-rate one does not, so it's the ROI interval we lead with.)

The dead zone: narrow home rest to exactly three days — the condition we originally published as Signal 1 — and the edge goes negative: 54.1% over 231 games sounds survivable on its own, but at the real closing moneyline that's −9.4% ROI, pooled across all four backfilled seasons (backtest_rest_signals.py). That's why we retired it instead of publishing it.

What we cannot explain: two days of rest wins money at real prices (+5.6% ROI, 509 games) and three days loses (−9.4%, 231 games) — both on samples large enough to take seriously. Beyond three days there's a pattern we're tracking, not a finding we're publishing: the 4+ day bucket swings back positive, but on just 80 games (+11.5% ROI, 95% CI −8.6% to +26.6%) — an interval wide enough to include a losing rate, so we are not calling it real. We do not have a clean mechanism for why the edge dips at exactly three days, and we are not claiming to know what happens after that either. We publish the two-day boundary because it is the one the pooled backtest actually supports at a sample size we trust, not because we have a tidy story for the whole shape — and we would rather say that than invent one.

Cancellation: when both teams are on a back-to-back, the rest difference disappears and no pick is published.

Honest limits: roughly 120–135 qualifying games a season means any single year proves very little, and every season so far has landed positive (+3.3% to +8.0%) rather than the flat or losing year a real long-run edge should eventually produce — treat that as good fortune so far, not a guarantee. Average prices have also swung more than we would have guessed: −186 in 2022–23 against −100 to −126 in the other three seasons, and we do not have a clean explanation for why. Every future result lands on this page, whichever way it goes.

What "Emerging" Means

Loading...

The backtest: Loading...

Comparison — away starts a backup instead: Loading...

The retirement test: if the live sample's first 100 games run net negative ROI, we retire Emerging Edge publicly — the same predetermined-threshold approach Signal 1 was retired under, not a judgment call made after the fact. Loading...

Live tracking: Loading...