How accurate is football match prediction data? Our settled record, in full

How accurate is football match prediction data? Our released record is 27 from 39 released picks, 14 April to 5 September 2026, and the full settled ledger is at /performance. That is the number that matters, because a fixture only reaches you if it clears the conviction threshold, and everything below it is filtered out before you ever see it.

This page is the long read underneath that headline, and it examines the first window in detail: the settled spring 2026 record, in which the fixtures the model surfaced landed 19 times in 22 from 14 April to 2 May 2026. The six picks released since, on 31 August 2026, are covered on the released board, where four of seven landed and the highest score of the round was one of the misses. Thirty nine settled picks is a small sample and the next thirty nine will not necessarily look like them. The split is the honest part: 19 from 22 across 14 April to 2 May 2026, then 8 from 17 since the new season opened on 31 August 2026.

What is 19 from 22 actually measuring?

It measures the fixtures we put in front of you. The model scores every match across the Championship, League One, League Two and the National League, but only those reaching 7.0 out of 10 are surfaced, so you get the games with the strongest data-backed case rather than a long list of noise. In the spring record, 22 fixtures cleared that bar and 19 of them landed.

That distinction is worth making, because plenty of services quote one impressive figure without ever saying what it is a figure of. Ours is the accuracy of the product itself. The full distribution, including every fixture you would never have been shown, is in the table below. Football is low-scoring and high-variance, draws swallow a large share of results, and one deflection routinely decides a game the data read correctly. So we publish all of it, misses included, and let the distribution speak.

What does our settled record actually show?

Between 14 April and 25 May 2026 the model scored 192 fixtures that reached settlement across the Championship, League One, League Two and the National League. Each was scored before kick-off across 8 data factors and given a conviction score out of 10. The results, by band:

Update, 10 September 2026, and it cuts against us. The table above is the spring window, 14 April to 2 May 2026. The 2026/27 season has run the staircase backwards so far. Released picks at 8.0 and above have landed 2 times in 6, and released picks below 8.0 have landed 6 times in 11, across 17 settled released picks between 31 August and 5 September 2026. Two rounds is not a sample and we are not presenting it as one, but the spring table on its own would imply the good run is typical, and on the evidence to date it is not. Our released record across both windows is 27 from 39 released picks, 14 April to 5 September 2026, with the full ledger at /performance.

One correction first, applied 13 August 2026. The bottom row previously read 94 and 22%. The model declined to name an outcome on 8 of the 94 fixtures it scored below 5.0, and a fixture it refused to call is not a prediction that failed, so those 8 sit outside the accuracy column. Called fixtures below 5.0: 86, of which 21 landed.

Two things about this table matter more than any single row. First, it climbs monotonically: every step up in conviction bought a real step up in accuracy, which is what calibration is supposed to look like. Second, nearly half the record sits in the lowest band. The model said "no strong case" about 94 of 192 fixtures, and those fixtures duly behaved like coin flips. A prediction system that never says "this one is genuinely unclear" is hiding that row, not avoiding it.

What does a low-conviction fixture look like in practice?

The final day of the National League season, Rochdale against York, is the record's best illustration. The model had Rochdale at home at 4.9, right at the top of the lowest band. York equalised in the 113th minute to draw 1-1 and take the title on the last kick of the season. A match the data called a coin flip was decided by one. That fixture lands on our miss record, and it is also the score working exactly as designed.

Did accuracy vary by division?

Substantially, and the differences are informative. At 7.0+ conviction the National League went 8 from 8, League One 7 from 9, the Championship 3 from 4 and League Two 1 from 1. More telling is how often each division produced a strong case at all: only 4 of 55 settled Championship fixtures reached 7.0 across 14 April to 25 May 2026, against 9 of 58 in League One over the same window. And below 5.0, League One predicted outcomes landed 1 time in 24 while the Championship's landed 8 times in 23. Put simply: the lower divisions separated more cleanly by data last spring, and the Championship's matches were genuinely closer. The model's conviction scores tracked that rather than forcing confidence where the data didn't support it.

What were the worst misses?

They stay on the record, because a record that only remembers its wins isn't a record.

The wildest was Mansfield 5-4 Cardiff, scored 7.2 for Cardiff, a nine-goal match that no factor model sees coming. Stockport at home to Port Vale was scored 7.0 and finished 1-2. Southampton against Bristol City was scored 7.1 and finished 2-2. Three misses from 22 fixtures at 7.0+ is the 19 from 22 above, seen from its uncomfortable side; that is the trade being made every time the model puts a high number on a match.

How should you use accuracy figures like these?

As calibration, not prophecy. A 7.0+ conviction score means the data-backed case for a particular match reading is strong; last spring that strength translated to 19 landed calls in 22, from 14 April to 2 May 2026, and the sample will keep growing every week of 2026/27. The full factor breakdown behind every score, and the live settled record it feeds, are on the results page. How each of the 8 factors works, with a fully worked example, is here: how the 8-factor conviction score works. For how the scores read on this season's contenders, start with the Championship and League One previews.

All figures from Lower League Analyst's settled-fixture record, 14 April to 25 May 2026, 192 fixtures. Accuracy means the model's predicted match outcome matched the settled result; it is a measure of prediction calibration and nothing else. Lower League Analyst is not a bookmaker and does not take bets. It is the research: the hard work done for you, the working shown, and the record published. No model can guarantee a result, and past performance is not a guide to future outcomes. 18+. Please gamble responsibly. begambleaware.org