Skip to content

Methodology

Calibration and tested matches

Two checks behind the track record. Calibration compares the probabilities we published with how often those outcomes happened. The tested matches show what our current method calculates for past matches — calculated afterwards, never published at the time and never counted in the track record.

All published predictions since the record began, settled at full time.

Calibration

When we say 60%, does it happen 60% of the time? Each bar is what actually happened in that probability band; the line is the probability we published.

Published predictions only

4,772 probabilities · all markets, not only picks

Our probability vs actual hit rate
  • Actual hit rate
  • Our probability
Our probability vs actual hit rate
Probability bandOur probabilityActual hit rateProbabilities
30–40%38.2%37.2%419
40–50%44.5%47.4%1006
50–60%54.7%54.6%2023
60–70%64.1%65.2%991
70–80%73.9%69.9%266
80–90%83.6%79.0%62
90–100%93.2%80.0%5 · small sample

Faded bands hold fewer than 30 probabilities — too few to read much into.

All tested past matches. The published picks are on the track record.

Tested on 17,490 past matches since

What our current method calculates for past matches, using only data that was available before each kick-off. These were calculated after the matches were played and never published at the time, so they are kept apart from the published track record.

See the tested matches

1,661 picks across 17,490 matches

Hit rate
82.1%
+14.9 pt vs baseline
1,661 picks
1X2
76.5%
+10.3 pt vs baseline
1,270 picks
O/U 2.5
100.0%
+29.7 pt vs baseline
391 picks