Every pick we have made, checked against the real result.

Nothing here is selected. A pick is published before kick-off, graded when the match finishes, and stays on the record whichever way it went. Losing picks are not removed, and the figure above moves on its own as results come in.

By sport

Football is the hardest of these and the one we cover most. Sports with fewer than 10 graded picks are left out rather than shown as a percentage of nothing.

Football70%2249/3231

Basketball66%377/568

Hockey60%129/215

Baseball56%45/81

Does the confidence mean anything?

The test that matters. When a pick is published at 70%, roughly 70 of every 100 such picks should win. The stated figure is on the left; what actually happened is the bar. Where they disagree, the confidence is wrong — and that is worth knowing about our own product.

Published at 50–59%27%3/11

Published at 60–69%59%87/148

Published at 70–79%73%555/765

Published at 80–89%89%572/646

Published at 90–99%92%312/338

Earlier picks, on the old scale

Predictions once carried a 1–10 rating rather than a probability. Those picks are still in the total above — removing them would flatter the record — but they are charted separately, because a “7” and a 70% are not the same claim and putting them on one axis would be dishonest.

Rated 5/1038%64/169

Rated 6/1053%564/1073

Rated 7/1067%465/695

Rated 8/1072%150/208

Rated 9/1079%26/33

How this is counted

One pick per match: The headline pick is the one graded. Publishing six markets and counting the one that landed is how a 50% record becomes a 90% claim.
Void picks are excluded, not counted as wins: A postponed match, or a pick we cannot check honestly, is left out of the count and left out of the count entirely. 124 of 4,219 fall into this.
Nothing is edited after kick-off: A prediction is written before the match and never revised. The record only means something if that holds.
Value picks are tracked separately: Where we disagree with the bookmakers, that call is graded on its own: 49% across 2,283 picks. A value pick is deliberately a longer shot, so it wins less often by design.

Today's predictions How picks are made

Does the confidence mean anything?

Published at 50–59%27%3/11

Published at 60–69%59%87/148

Published at 70–79%73%555/765

Published at 80–89%89%572/646

Published at 90–99%92%312/338

Earlier picks, on the old scale

Rated 5/1038%64/169

Rated 6/1053%564/1073

Rated 7/1067%465/695

Rated 8/1072%150/208

Rated 9/1079%26/33

How this is counted

One pick per match

The headline pick is the one graded. Publishing six markets and counting the one that landed is how a 50% record becomes a 90% claim.

Void picks are excluded, not counted as wins

A postponed match, or a pick we cannot check honestly, is left out of the count and left out of the count entirely. 124 of 4,219 fall into this.

Nothing is edited after kick-off

A prediction is written before the match and never revised. The record only means something if that holds.

Value picks are tracked separately

Where we disagree with the bookmakers, that call is graded on its own: 49% across 2,283 picks. A value pick is deliberately a longer shot, so it wins less often by design.