scienceResearch preview · Frozen before kickoff

The Sirius record — every forecast, graded

Sirius reads each game from measured evidence and publishes three chances — home, draw, away — frozen before kickoff. This page grades all three against what actually happened, misses included, plus the games it honestly sat out.

Updated as games finish · last graded game 15 Oct 2026

Games graded
2129
6387 claims — every forecast grades all three chances
Typical chance on the result
36.9%
market consensus, same games: 37.9%
Average miss
1.5 pts
distance between said and happened, per claim
No forecast
567
not enough track record — an honest sit-out, kept on the record

When we say 60%, does it happen 60% of the time?

Every claim we’ve published, sorted by the chance we gave it. Matching pairs of bars are the whole point of a probability forecast — the grey bar is the promise, the filled one is what football delivered.

0–10%
242 claims
we said
7%
it happened
5.1%11.9%
7.9%
10–20%
915 claims
we said
16%
it happened
14.3%19.2%
16.6%
20–30%
2203 claims
we said
25.6%
it happened
25.6%29.4%
27.5%
30–40%
1288 claims
we said
34%
it happened
30%35.1%
32.5%
40–50%
677 claims
we said
44.2%
it happened
37.7%45.1%
41.4%
50–60%
561 claims
we said
54.7%
it happened
49%57.2%
53.1%
60–70%
258 claims
we said
65%
it happened
59.5%71%
65.5%
70–80%
166 claims
we said
74.5%
it happened
66.9%80.2%
74.1%
80–90%
74 claims
we said
84.5%
it happened
72.2%89.4%
82.4%
90–100%
3 claims
still collecting
we said
90.5%
it happened
43.8%100%
100%

On each “it happened” bar, the two thin marks bracket where the true rate could sit by chance alone — flip a fair coin 20 times and you can easily get 13 heads, not 10; small samples wobble. A gap between said and happened only starts to matter once “what we said” falls outside the marks, and more claims pull the marks closer together.

Is this better than just guessing?

Four ways to forecast the same 1323 graded games, scored with the forecast-error score (statisticians’ Brier): 0 is perfect, 2 is confidently wrong — lower is better.

← betterMarket consensus0.577the market’s average, margin removedSirius0.589our model — the market is never an inputLong-run averages0.648always forecast home 44% · draw 25% · away 30%Pure guessing0.667an even 33% on every result

The colored stretch of the line is the ground we’ve covered — 86% of the distance from pure guessing to the market consensus. Scored only where the consensus was saved at least 30 minutes before kickoff. Across all 2129 graded games — market consensus saved or not — we score 0.596; the head-to-head stays on the shared 1323, where both reads cover the same games. Every term on this page, explained in the glossary. How these forecasts are made — the full methodology.

For what it’s worth on most-likely calls alone: ours landed 673 of 1323, the market’s 696. We grade the whole distribution, not the call — that single number hides everything this page is for.

Every forecast

Row tint — where the result ranked in our forecast:our top chanceour secondour third

“We gave it” is the chance the frozen forecast put on the result that then happened — 100 would be clairvoyance, and the thin mark on each bar is this record’s typical 36.9%. A 42% result landing is not a triumph and an 18% one is not a disaster; the bands above are where the honest judgement lives.

Across 6387 graded claims our number misses what happened by 1.5 points on average. In all 9 ranges with enough claims to judge, what we said sits within what chance alone allows — inside the thin marks on the bars above.

A fair comparison needs the market consensus saved before kickoff, and we have it for 1323 of the 2129 graded games. On those, the market has been closer to what happened than us — it misses by 1.3 points on average, we miss by 2.1. We are behind it, and this line will say so for exactly as long as that stays true.

Worth weighing before you lean on any of it: only 297 of the 2129 graded games come from the top European leagues Sirius is tuned for — the remaining 1832 were earned in other competitions. This caveat keeps count, and retires once the tuned-for leagues carry a real share of the record.

Complete record since 7 July 2026, updated daily as games finish. Every number was written before kickoff and frozen; nothing is removed, nothing is regraded. The record so far spans 38 competitions and 22 engine versions; the largest competitions are Conference League (262 games), MLS (217) and Série A (122).

Open data. The figures on this page are Oddsivio's own record and are free to reuse under CC BY 4.0. Credit Oddsivio and, if you can, link to this page — a nofollow link or a plain-text credit is fine. The underlying match results and other third-party data are not included; see our terms.

We use cookies to measure how the site is used so we can improve it. Change your choice anytime via “Manage cookies” in the footer. See our Privacy Policy.