PSG are thirteenth. The model still makes them champions.

The table says one thing, the model says another. After three matchdays, Paris Saint Germain sit 13^th on 8 points, behind Le Mans and Auxerre. Monaco lead with nine points from nine. And yet, across fifty thousand simulated seasons, it is PSG who lift the trophy in 60% of them — against 15% for Monaco.

Three matchdays weigh almost nothing

The model is fitted on more than two thousand Ligue 1 matches — seven seasons. The three matchdays played represent a little over one per cent of that. Even giving far more weight to recent matches — which the model does, with a half-life calibrated on the data itself — the information carried by the start of a season stays small against years of dominance.

This is not an arbitrary choice. The half-life used for Ligue 1 is the one that, across past seasons, best predicted the following matchday. A model that forgot faster would be swept along by the first accident; a model that forgot more slowly would miss the real turning points. This one sits where the data points.

What the model sees that the table does not show

The table adds up points. The model estimates two quantities per team: what it scores and what it concedes, adjusted for the quality of its opponents.

On that ground, PSG remain first in the league. Their attack index is 1.80 — so they score about 80% more than an average side against an average defence — and their defence index 0.83, close to twenty per cent fewer goals conceded than average. No other side in the league combines the two at that level.

So why only two points?

Because three matches are three matches. A side that wins 65% of its games loses or drops points in one in three, and sometimes the three fall in a row. The model knows this is common: across fifty thousand simulated seasons, there are plenty where PSG start badly and finish first.

What costs PSG is not a drop in level — it is being seven points behind Monaco after three matchdays. A gap closes mechanically if the level is higher; there simply have to be enough matchdays left, and thirty-one remain.

Where each team is heading

Average expected position at each matchday. Left of the marker, what actually happened; right of it, the projection.

Olympique Lyonnais4.8^eParis Saint Germain1.4^eMonaco3.3^e

What the projection says

At the finish, the model puts PSG on 68.9 points on average, against 61.3 for Monaco. The gap in level is such that it absorbs the current deficit — but not to the point of making it a certainty: four simulated seasons in ten end without a Paris title.

This is where the difference between a tip and a probability plays out. Nobody here is saying PSG will be champions. The model says they are champions in a majority of possible futures, and that the remaining minority is not negligible.

What would really move the figures

Two things, and only one of them is happening.

The gap can become unbridgeable. Every matchday without points costs more than the last, because fewer matches remain to come back. That is the dominant mechanism today.

The level may genuinely have dropped. A departure, a long injury, a change of system: the model will register it eventually, but slowly, and only through results. It does not read team sheets, knows nothing of transfers, and will never know a player is missing until the side has started scoring less.

For now, the two effects do not weigh the same. The model has barely revised its estimate of Paris's level — it simply observes that PSG have dropped points, not that they have become weak.