Why we don't score fixtures until four games into the season

Football prediction models are at their least reliable in the opening weeks of a season, and ours is no exception, so it waits. Until both clubs in a fixture have played 4 league matches, Lower League Analyst scores the game internally and publishes nothing. This week that meant scoring all 60 opening fixtures across four divisions and releasing none of them.

Why do football prediction models struggle early in the season?

The consensus answer is the right one: there is almost nothing current to measure. Pundits and stats sites broadly agree that the early-season table misleads, and the reason is structural. Squads turn over every summer, promoted clubs arrive with no record at the new level, and last season's form belongs to a team that no longer exists in quite the same shape. Most of the 8 factors our model scores, form, home and away splits, shots, goals and clean sheets among them, reset to zero in June.

What survives the summer is thin. The main carry-over is head-to-head record, and in our settled record head-to-head was the weakest separator of all 8 factors: across 175 settled fixtures carrying a head-to-head score, the gap between its average score on correct calls and on misses was just 0.15, the narrowest of any factor we measure. The one input that survives the summer is the one that tells you least.

What does the model actually see in August?

Here is the live picture, pulled from our own scoring engine on 6 August 2026:

The engine has scored all 60 fixtures from 8 to 22 August across the Championship, League One, League Two and the National League. Every one of the 60 is flagged insufficient data, and none will be published. 54 of them rest on exactly one factor, head-to-head record. The other six carry no factor at all, because the clubs involved have no recent meetings: Notts County v Leicester, Barnsley v Bromley, Middlesbrough v Lincoln, Cardiff v Wrexham, Carlisle v Worthing and AFC Hornchurch v Kidderminster.

Across the 184 settled fixtures the model scored in last season's run-in, the average fixture was built on 6.8 factors. The opening card averages 0.9. That is the whole argument in one number: in August the model holds about one eighth of its normal evidence, and it would be dishonest to pretend the output means the same thing.

There is a neat irony in the zero-factor list. The televised League One opener, Notts County v Leicester, is the fixture the model currently knows least about in the entire division.

What would happen if we published anyway?

One fixture makes the case on its own. Scunthorpe v Yeovil on opening day currently scores a raw 7.3, which in-season would put it on the conviction board. It is staying unpublished, because that 7.3 is built on a single input, Scunthorpe winning 3 of the last 5 meetings, and nothing else.

Compare what a 7.0+ score meant when the model had full data. Across the released record, 22 picks from 14 April to 2 May 2026, fixtures scoring 7.0 or higher landed 19 times in 22. Most carried a full or near-full factor set. Two carried none at all, which our own gate should have caught and did not, and we are fixing that. A one-factor 7.3 wears the same badge with none of the evidence behind it. Publishing it would trade the meaning of the number for a week of content.

The reverse error is uglier. Carlisle v Worthing is currently stored at 0.0, not because the data condemns Carlisle but because Worthing are newly promoted and the pair have no head-to-head at all. A published 0.0 reads as a verdict. It is actually a shrug, and the difference between those two things is exactly what a conviction score exists to protect.

How much data does a football prediction model need?

Our threshold is 6 matches, applied per fixture: both clubs need 6 league games of the current season behind them before the model scores the game properly. That is roughly the point where a home record stops being a single result dressed as a trend, and where shots and goals data covers enough different opponents to mean something. It is a floor, not a magic number; the reads keep firming up for weeks afterwards.

The threshold is uniform across all four divisions. The National League simply reaches it first in calendar time, because its clubs routinely play twice a week.

When do the first conviction boards arrive?

National League fixtures should be the first to qualify, in late August. The Championship, League One and League Two follow in early September. Until then, the pre-season reads live on the four division previews, all built on last season's settled record and labelled as such: Championship, League One, League Two and National League.

Every score the model did publish last season is settled and on the record: how accurate is football match prediction data. The method behind the number is here: how the 8-factor conviction score works, and the dashboard will show every fixture, all 8 factors and the working the moment the threshold is met. What we will not do is publish a number in August that we would have to disown in September.

Opening-card figures pulled live from Lower League Analyst's scoring engine on 6 August 2026 (60 fixtures, 8 to 22 August). Run-in figures from the settled record, 14 April to 25 May 2026. 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

Updated 10 August 2026: the summary figure for the settled spring record's factor average is corrected to 6.8 factors per fixture. An earlier version read 7.1.

Update, 10 September 2026: the threshold moved from four matches to six

This page was written when the gate stood at four league games. On 6 September 2026 it was raised to six, and a price cap was added alongside it. A fixture now reaches the board only when it clears the release threshold, both clubs have six league games of the current season behind them, and the price on our call sits at or below 2.20.

The reasoning for six is the reasoning for four, applied to what the first month of the season actually showed. At four games a home record rests on two home fixtures, which is not a home record. Six gives three, and it also lets the form factor drop a club's opening two results, which in this division are the least representative games of the year.

The price cap is a separate constraint and it is not about accuracy. It caps how far out the market has to be before we are willing to put a fixture in front of a subscriber, and on the 12 September 2026 card it withheld three fixtures that had already cleared both other gates. Two fixtures were released that day. Three were withheld on price and two on the games-played gate, so the cap withheld more of that card than the gate did.

The URL of this page still says four. We have left it alone deliberately, because changing it would break every link pointing here. The rule is six.