A backtest is the cheapest truth in betting: take a rule — “bet home underdogs off a loss when the total is under 45” — replay it over past seasons, grade every bet it would have made, and read the record. Done honestly, it kills bad ideas for free, before they cost real units. Done carelessly, it manufactures confidence in noise — which is worse than no test at all, because now the bad idea has a chart.
The method, in five rules
- 1. Freeze the rule before you test it. Write the exact conditions down first. Rules edited mid-test to improve the record aren't being tested — they're being fitted.
- 2. Grade every matching game. Skipping the games that “felt different” is selection bias, and it will conjure an edge out of a coin flip. If the conditions matched, the bet counts.
- 3. Score against break-even, not 50%. At −110 the bar is 52.4%. A 55% rule is real money; a 51% rule is a donation. The break-even calculator converts any price into the bar your record must clear.
- 4. Demand a real sample. 14–6 over half a season is a coin doing tricks — streaks like that happen constantly by chance. A few hundred matched games across several seasons is where evidence starts.
- 5. Read the record per season. A rule that crushed 2022 and has faded every year since isn't an edge — it's a market inefficiency that got priced in. You want the profit spread across seasons, not concentrated in one.
The trap that catches almost everyone: overfitting
Add enough conditions and history will always confess. A 10-condition rule that went 68% over five seasons sounds like a goldmine; it is almost always a description of the past wearing a lab coat — each extra condition carved away the losses until only wins remained. The tell: tightening a condition's threshold (from “3+ rest days” to “exactly 4”) because it improves the record. Real edges are robust — they survive small changes to the rule. Fragile records are fitted, and fitted records collapse on contact with next week. Two to four defensible conditions beats ten clever ones, every time.
The problem nobody warns you about: the data
An honest backtest needs game results, closing lines, rest and schedule context, and player availability — per game, per season, per league. Assembling that in a spreadsheet is weeks of work, and historical odds are the expensive part. This is the grunt work Covrr exists to delete: the model builder backtests your conditions against five seasons of NFL, NBA, MLB & NHL games — with real odds, per-season breakdowns, and ROI — in seconds, free, before you've made an account.
What to do with a rule that passes
Treat the backtest as the rule's contract: it earned its stakes at a measured win rate, so judge it on that sample going forward — not on last night. Convert the record into an edge with the EV calculator, size stakes with units or fractional Kelly, and let the rule keep working. The full pipeline from idea to deployed model is in our step-by-step guide.