How We Verify Results
Why the numbers on this site can be checked rather than taken on trust.
Predictions are recorded before kickoff
Every prediction is written to the record before the match starts, together with the selection and the price available at that moment. A prediction cannot be created for a match that has already begun, and no prediction is added after a result is known.
This is the single most important property of the record. A track record assembled after the fact is not evidence of anything; one written in advance is.
Records are timestamped and not edited
Each prediction carries the time it was created and the model version that produced it. Once written, the authorship and creation time of a prediction cannot be changed — this is enforced by the database, not by convention.
Settlement is automatic
When a match finishes, predictions are settled automatically against the final result: both teams to score is settled on the goals, over/under on the total, match result on the score, corners on the corner count. Settlement runs on a schedule and is not a manual judgement. Matches that are postponed or abandoned are voided rather than counted as wins or losses.
Losses are retained
Losing predictions stay in the record permanently. The public results page reports win rate, ROI and P&L including losing periods, and shows negative figures when the period was negative. Nothing is filtered to show only winners, and no minimum-performance rule hides a poor result.
Retired traders stay in the historical record
The AI Trader roster shows 7 active and 2 retired traders. Retiring a trader does not remove its history: its settled bets, win-loss record and profit or loss remain published. Removing a retired trader would quietly delete part of the evidence — often the part with the largest sample.
How ROI and P&L are calculated
One canonical definition is used everywhere on the site:
- Settled bets — predictions with a final outcome of won or lost. Voided and pending selections are excluded from performance figures.
- P&L — for a winning selection, the price minus one stake unit; for a losing selection, minus one stake unit. Summed across settled bets.
- ROI — P&L divided by total amount staked, expressed as a percentage.
- Win rate — winning selections divided by settled selections.
The headline public settled count excludes lay-model selections, which are recorded separately because their staking is not comparable on the same basis.
Sample size matters
Every figure is published with the number of settled selections behind it. Small samples are not meaningful: a handful of bets can produce a large positive or negative ROI purely by chance. Figures on this site are historical records, not forecasts, and past performance does not indicate future results.
What this does and does not prove
The record shows what was predicted, when, at what price, and what happened. It does not, on its own, prove that Punter Insight's selections beat the market over the long run — that is a harder claim requiring a larger and more carefully controlled body of evidence, and Punter Insight does not currently make it.
See also the methodology and how it works.