Why a good pick still loses: probability, variance and streaks
Published October 19, 2026
A prediction is a probability, not a promise. A pick given 65% is expected to lose about 35 times out of 100 — and when it does, nothing went wrong.
What a probability promises
When iPick gives a pick 60%, it means that out of many games like this one, the pick should win about 60 and lose about 40. The test of a probability is not whether this game was won, but whether, over many games, the 60% picks won about 60% of the time. That property is called calibration.
Over the last 30 days to October 14, 2026, iPick's 519 published pre-game picks said, on average, that they would win 65.3% of the time; they won 65.9%. That is what a calibrated record looks like.
Three levels, three different promises
To make this visible, every pick on iPick is shown in one of three levels of its own probability. Over the 536 graded picks published between August 20 and October 16, 2026:
- Strong (65% or more): 164 of 207 won — 79%.
- Lean (55% to 65%): 114 of 184 won — 62%.
- Toss-up (under 55%): 75 of 145 won — 52%, about a coin flip, as the number said.
Most of the picks that are seen to “lose a lot” are Toss-ups. They are shown as such so that nobody mistakes a 52% for a sure thing. The live figures are on the Track record page.
How much luck is in a short run
Even a predictor that is right 65% of the time in the long run will show very different results over short stretches. With pure chance alone, a true 65% record lands, 95 times out of 100, somewhere between:
- 47% and 80% over 30 games;
- 56% and 74% over 100 games;
- 60% and 70% over 400 games.
So a bad week — say 13 right out of 30 — is not evidence that something broke. In mid-October 2026, iPick went 13 for 30 over three days; on the same 30 games the sportsbook favourite went 14 for 30. Upsets hit everybody at once.
Streaks are normal
At a 65% hit rate, three losses in a row happen about 4% of the time on any given run of three picks — over a season of hundreds of picks, many times. Long streaks of wins happen too, and are just as uninformative.
The only way to tell skill from luck is a large sample and a fair comparison: see How to judge a sports prediction.