How to judge a strategy's record: sample, CLV, ROI

Before any other number, read how many graded picks sit behind it. A strategy at 12-3 has a striking win rate and almost no evidence. Records swing widely over dozens of bets and settle only over hundreds. A record under a hundred graded picks is a reason to keep watching, not a reason to follow.

Check the window as well. An all-time record and a last-30-days record can tell opposite stories. Neither is the true one. The long record says more about the process; the short one says more about the present, with far less evidence.

Closing line value, or CLV, measures how often a strategy's picks got a better price than the market closed at. It is steadier than win rate because it does not depend on who won. A strategy with positive average CLV over hundreds of picks is beating the most efficient price the market produces. A strategy with negative CLV that is still up is likely to give it back.

CLV is not proof on its own. It can be inflated by betting into soft or thin markets whose closing lines are weak, and it says nothing about results in markets that close badly. Read it as the best single piece of evidence, not the only one.

ROI is profit divided by the amount staked. Units are profit in stakes of one. Together they tell you what following the strategy would actually have earned or lost at the published prices. Win rate alone cannot, because a 45% record on underdogs can earn more than a 58% record on heavy favourites.

Always read units with their sign. A record that leads with its win rate and leaves out a negative unit total is hiding the part that matters.

We report results as they happened. We do not adjust records for close losses, bad bounces or anything else, and we do not rank strategies on what they should have done. A loss counts as a loss. If you judge records elsewhere, be wary of any that are adjusted: an adjustment is a choice, and it is rarely made against the person making it.

A rough guide comes from the arithmetic of coin flips. Over 100 bets at a true 50% rate, the win rate falls between about 40% and 60% most of the time. Over 400 bets, between about 45% and 55%. Over 1,600, between about 47.5% and 52.5%. A record has to clear those bands by a clear margin before its win rate says much.

At -110, break-even is 52.38%. A strategy winning 55% over 100 picks is inside the band a break-even strategy would produce. The same 55% over 1,600 picks is outside it. That is the difference between a record and evidence.

Cherry-picked windows: a service that shows its best 30 days, or starts its record after a losing stretch. Fix it by asking for the full history and its start date.

Mixed markets: a record that blends one strong market with several weak ones, or that changes which picks count. Fix it by reading each league and market on its own.

Missing prices: a win-loss record with no prices attached. A 60% record on -200 favourites loses money. Without the prices, a win rate cannot be judged at all.

The Strategies board lists each model by league and market, with its graded record, units, ROI, average CLV, sample size, maximum drawdown, profit factor and a Brier score for calibration. A strategy score combines CLV, calibration, ROI and sample size, and it is not shown under thirty graded picks, because a score on a handful of results would be noise.

Filter to the league and market you care about and set a minimum number of picks before you sort. Sorting the whole board by ROI with no minimum puts the smallest samples at the top. Every pick is graded automatically against the final result and the closing line, as our grading page describes.

One limit to know: underperforming models may be withheld from public view, and the board says so. The pooled record on the records page is built from the models shown. Read it with that in mind.