How to Verify a Sports Prediction Track Record
A checklist for evaluating win-loss claims, timestamps, sample size, pushes, canceled picks and the difference between a result and a prediction.
1. Look for a record made before the event
The essential evidence is the original selection recorded before the game began. A recap posted afterward can confirm the outcome but cannot establish that the author made the prediction in advance. Useful audit details include event ID, matchup, league, pick type, quoted selection, original issuance time and scheduled start time. Independent preservation of that timestamp adds confidence.
2. Ask exactly what the reported record counts
A headline such as '8–4' is meaningless without a defined set of picks and dates. Does it include moneylines and spreads together? Is it season-to-date, lifetime or a rolling 30-day window? Are multiple selections on the same game counted as separate predictions? Clarify the denominator before comparing win rates across sites. If the method changes, the history should say so.
3. Losses, pushes, voids and pending outcomes matter
A win rate based only on winning screenshots is selection bias. Check that graded losses remain visible. A push usually has different settlement treatment than a loss, while canceled or ungraded games should not quietly disappear from the audit log. Read the publisher's stated grading policy and check examples against official results. Some markets have overtime and other special rules.
4. Distinguish verified model picks from owner-attested history
An owner can honestly remember and document an older pick after the event. That may be useful background, but without an independently preserved pregame timestamp it should not inflate the model's verified record. The historical note should show its source and provenance alongside the final score, and live-model percentages should use only their defined timestamped population.
5. Consider uncertainty and sample size
Even a truthful short winning streak is weak evidence of predictive skill. A small sample is noisy, and different odds imply different baseline expectations. Percentage correct does not by itself describe profitability or the quality of the published price. The responsible presentation includes the number of predictions, a clear period, and disclosure of uncertainty instead of guarantees.
6. A simple public audit routine
Pick a few random winning and losing entries. Verify the game, date and posted selection, compare the original timestamp to kickoff, and independently check the result. Flag missing evidence rather than filling it in. Sports Genius separately labels its owner-attested history and its timestamped live-model record so these two types of evidence cannot be confused.