The Model Has Auxerre at 44 Per Cent and I Am Not Entirely Sure It Has Gone Far Enough

When the market goes quiet, the model has to speak for itself

There is a strange honesty to a fixture with no prices attached. Every other week of the season I am working the seam between what the model thinks and what the market thinks, hunting the gap, arguing that the number on the screen has drifted away from the number in the machine. Not here. No live market is attached to Auxerre against Stade Brestois 29 yet, which means the Lemeister forecast of Auxerre 44 per cent, draw 28, Stade Brestois 28 stands naked and unhedged. There is nothing to agree with, nothing to argue against, just a single clean read that says the home side are likelier than not to avoid defeat and likelier than anything else to win.

That is a curious sentence to write about a team sitting in the middle of the French top flight, but the numbers behind it are worth sitting with. MeisterIQ, our conviction measure, comes in at 53 out of 100. That is not a shrug and it is not a roar. It sits in the honest middle: the model has a lean, it believes the lean, but it is telling us not to treat this as one of those afternoons where the result is effectively pre-written. A 44 per cent home win with a conviction of 53 is the sound of a machine that has looked at everything and concluded that the most probable single outcome is the one it has given you, while also conceding that the two other outcomes together carry 56 per cent of the probability between them.

That is the whole tension of this match in one line. Read 44 per cent carelessly and you think "Auxerre are clear favourites". Read it properly and you think "the home side are marginal favourites in a game that could go anywhere". The gap between those two readings is where the entire preview lives.

Auxerre, by the cold light of the archive

Here is the thing that makes Auxerre a genuinely interesting model subject rather than a throwaway name. Across the Lemeister archive there are 1229 recorded Auxerre matches, running from 1993 all the way to this season in 2026. They have won 461 of them, drawn 354, lost 414. That is a win rate of 38 per cent, and if you do the arithmetic on the rest it is a side that has spent three decades being roughly two parts winner, two parts loser and one part stalemate.

The goals column is even more revealing. 1533 scored, 1386 conceded. A positive goal difference of 147 across more than twelve hundred matches. Over a stretch that long, that is not a fluke and it is not a golden generation either. It is the profile of a club that has generally been a little too good to be bad and a little too limited to be great. They score slightly more than they ship, they win slightly more than they lose, and they have built an entire identity on being a nuisance to bigger clubs rather than a threat to them.

I like the 38 per cent figure because of how unglamorous it is. It tells you Auxerre are not a side who arrive at a home fixture with any divine right to win it. But 1229 matches is a very long memory for a model to lean on, and the goal totals tell you the underlying quality has usually been there. When Auxerre get the conditions they want, they convert. When they do not, they have historically been just as likely to drop two points as take three and that, more than anything, is why the model caps them at 44 rather than pushing them towards the high fifties where a genuinely dominant home favourite would sit.

The uncomfortable symmetry of a draw at 28 per cent

Draws get ignored in previews because they are boring to write about, and then a third of the season disappears into that blind spot. Here the model hands the draw 28 per cent, which is a couple of points above what a flat 1X2 split would give it and meaningfully above what a real mismatch would produce. This is the model quietly telling you it expects a tight, low-margin game in which neither side manages to establish the kind of control that turns three points into a formality.

And look at the shape of it across all three outcomes. 44, 28, 28. The away win is priced identically to the draw. That is not a throwaway detail. It means the model sees no real distinction between "honours even" and "Brest nick it", which is a polite way of saying the away side are live and the home advantage is doing some work but not a great deal of it. If you take the draw and the away win together you get 56 per cent, a majority, sitting on the two outcomes Auxerre do not win. That is the single most provocative number in this whole forecast and I would rather say it plainly than dress it up: by this model, the likelier collective outcome is that Auxerre do not win.

That is not a prediction. It is a probability distribution, and anyone who reads 56 per cent as "back the away side or the draw" has misunderstood what a forecast is. But it is the honest read of a fixture the model has declined to treat as a home banker, and it is why the conviction number sits at a moderate 53 rather than soaring into the seventies. The model has conviction in its lean. It does not have conviction in a rout.

What a number like this does and does not tell you

I want to be very careful here, because this is the part of the job where people stop listening and start reaching for their phones. A forecast is not advice. Nothing I write about the 44, the 28 or the 28 is a recommendation, a tip or a steer, and I am not going to pretend otherwise by dressing a probability up as a certainty. The model has no idea what will happen on Sunday afternoon. It has an opinion, expressed in percentages, and it is telling you how strongly it holds that opinion through the 53 out of 100 conviction score.

What a number like 44 actually gives you is a frame. It tells you that if this exact fixture were played a hundred times, the model's central expectation is that Auxerre win forty-four of them, that twenty-eight end level and that Brest win twenty-eight. Whether that distribution turns out to be well calibrated is a question we cannot answer from a single Sunday in September. What we can say is that the model is not being lazy. A lazy model would have looked at home advantage, shrugged and handed Auxerre something in the fifties. Ours has done the work and pulled them back down to 44, which means it has found something in the history, the form profile or the matchup that says "this is closer than the venue suggests".

The archive numbers give us a clue as to why. A 38 per cent all-time win rate is a model input that actively resists home-team inflation. You cannot feed twelve hundred matches of two-parts-winner football into a machine and expect it to suddenly declare a fortress. Auxerre's long-run record is the record of a club that wins more than it loses but rarely dominates, and the model is respecting that history rather than overriding it because of the badge or the kickoff time.

Brest at 28 and the trap of the ignored underdog

There is a school of thought, and I have been guilty of it myself, that treats any away side under 30 per cent as an afterthought. That is a mistake here. Brest at 28 is the same number as the draw, and the draw is the outcome the model has decided is genuinely plausible. If the draw deserves respect, so does the away win, because the model has bundled them together as near-twins.

This is the fixture where a contrarian has to be honest about what he is and is not seeing. My usual game is finding the side the model rates higher than the market does and making the case that the market has been slow. There is no market here to beat. So the discipline shifts. Instead of hunting an edge in points, I have to interrogate the model itself and ask whether 53 conviction is enough to trust the shape of this thing. And my read is that the shape is the story. The 44, the 28, the 28 does not describe a favourite and an underdog. It describes a game between two sides the model regards as closer than the league table and the venue might imply, with a modest nudge towards the hosts for the simple human reason that they are at home and someone has to be slightly likelier than the others.

That is a less exciting conclusion than a bold call on a long shot, and I am comfortable with it. The most useful thing a forecast can do on a Sunday with no prices attached is to tell you, without any salesmanship, how open the game really is. This one is open. The model has Auxerre ahead, Brest and the draw level behind them, and a conviction score that stops short of certainty. Sixty per cent of the fun of a match like this is that the model, for all its twelve hundred Auxerre matches and its careful arithmetic, cannot tell you which way Sunday falls. It can only tell you that it is closer than it looks.