The record

Every settled prediction

Loading the published record…

The record

Every settled prediction

63.9%of 6,073 settled high-confidence predictions (60%+) came in.

6,073 settled at 60%+ confidence · last 90 days · model: composite

Across all 40,607 settled predictions — including the riskier, lower-confidence ones — 48.6% came in.

This is every prediction our model has made and had graded in the last 90 days — the losses included. The headline figure is the high-confidence segment: picks the model rates at 60% or more, the ones we promote. That threshold is fixed, not tuned, and the rate across every settled prediction sits directly beneath it. Nothing is removed from the archive below, and the figures move as results come in.

A hit rate is not a promise about your next bet. The model states a probability for each outcome, and a probability is a claim about the long run. The chart below is the more honest test: it asks whether the numbers we state actually hold up, which is a harder question than whether we were right more often than not.

Where it comes from

Every split carries its own sample size. A model can post a respectable overall number while being near-useless on one outcome — and the draw row below is exactly that case, on a sample far too small to conclude anything from.

All picks

48.6%

40,607 settled

home

53.3%

22,432 settled

draw

20.0%

30 settled

away

42.8%

18,145 settled

Does 60% mean 60%?

Each point is a band of stated probability. Its height is how often those predictions actually came in. The dashed line is perfect calibration — a point on it means the number meant what it said.

Above the line is under-confidence: the outcome happened more often than we claimed. Below is over-confidence, which is the direction that costs money.

Our stated probabilities currently sit an average of 5.9 points away from what happened, across 40,607 settled predictions. That is a measure of how well-calibrated the model is — it is not an accuracy figure, and it should not be read as one.

The bars beneath the axis are the sample size in each band. They matter: a band holding two predictions and one holding two thousand draw the same dot, and only one of them means anything.

Profit and loss

Not published yet — but the clock has started.

A return figure needs the price a prediction was available at when we published it. We now freeze that price against each prediction at the moment it goes out, together with the bookmaker and the time the price was quoted. A profit figure will appear here once enough of those have been settled to mean anything.

Nothing published before that change can ever be included, and it is worth being exact about why: we only keep bookmaker odds for three days, and each refresh overwrites the previous price rather than adding to it. The prices our older predictions were available at are not stale in our records — they are gone. Reconstructing them from today’s odds would produce precisely the flattered number this section exists to refuse.

Not every prediction will carry one. Our feed prices a match-winner market on roughly two thirds of fixtures — 127 of 200 on the day this was measured — and a prediction with no published price is left out of the calculation rather than counted as a bet that returned nothing.

The archive

40,607 settled

Newest first, wins and losses together, unfiltered.

All results Won Lost

Settled prediction archive, showing 25161 to 25200 of 40,607 results
Settled	Fixture	League	Our pick	We said	Result	Outcome
24 Sept	LNZ Cherkasy U19 Karpaty Lviv U19	U19 League	Karpaty Lviv U19	42.8%	LNZ Cherkasy U19	Lost
24 Sept	Vissel Kobe Sagan Tosu	Emperor Cup	Vissel Kobe	40.0%	Vissel Kobe	Won
24 Sept	Machida Zelvia Tochigi City	Emperor Cup	Machida Zelvia	41.6%	Machida Zelvia	Won
24 Sept	Kyoto Sanga Fujieda MYFC	Emperor Cup	Kyoto Sanga	40.3%	Kyoto Sanga	Won
24 Sept	Kashiwa Reysol Imabari	Emperor Cup