Ligue 1 · France · Regular Season - 4
Too close to call — we gave this result 26%
This verdict goes into the public record unchanged; we do not filter out the ones we get wrong. The model's track record
| Paris FC | Lyon | |
|---|---|---|
| 6 | Shots | 5 |
| 4 | Shots on target | 1 |
| 3 | Shots inside the box | 3 |
| 0 | Goalkeeper saves | 4 |
| 4 | Corners | 3 |
| 2 | Offsides | 1 |
| 56% | Possession | 44% |
| 590 | Passes | 459 |
| 534 | Accurate passes |
Formation: 3-4-2-1 · Manager: L. Rosenior
Provider data — may change until the line-ups are announced. Last updated: Thursday, 17 September 2026 at 06:10.
There's no card data for this referee from before this match yet. We don't invent an average; it will appear here once the data exists.
This is a prediction, not a guarantee; a probability is not a certainty.
These sentences are built by a fixed template; every number in them comes straight from the model's computation, none invented while the text was written.
When the model has said 35–40% in the past, 1,562 of 4,187 predictions were correct (37.3%). Accuracy record →
| 392 |
| 19 | Fouls | 13 |
|---|
| 2 | Yellow cards | 1 |
|---|
| 0 | Red cards | 0 |
|---|
The measurements our data provider sent for this match. A row that was not measured shows “—”; writing 0 into an empty field would claim we measured it and found none.
Formation: 4-3-2-1 · Manager: Paulo Fonseca
Lineups are announced about an hour before kick-off and can still change until the first whistle. The list is shown exactly as the provider sends it and does not feed the prediction. Last updated: Saturday, 12 September 2026 at 18:38.
When the model has said 65–70% in the past, 575 of 859 predictions were correct (66.9%). Accuracy record →
These three values add up to 200%, not 100% — and that is not a mistake. The options are not mutually exclusive; every outcome is counted in two of them. For the same reason 1X plus X2 exceeds 100% by exactly the probability of a draw (25.8%).
First-half double chance
These values are not a fraction of the full-match prediction; they come from a separate model trained on the first-half scores of 1326 matches. Its edge over the base rate is markedly smaller than the full-match model's — the measured figure is on the accuracy record. Read them more cautiously. First-half goal markets are not published because they failed to beat the base rate.
Fouls
Corners
These figures are set to hold in at least 60% of matches; in the rest the count comes in below them. No upper bound is stated — we make no "at most" claim. Yellow cards were measured and are not published: there our model did not beat the historical average.
When the model has said 60–65% in the past, 1,139 of 1,798 predictions were correct (63.3%). Accuracy record →
The league's historical "3+ goals" rate is 54.6%. For this match the model gives a value above that baseline — and in backtesting the model's edge was confirmed only in this direction.
Expected goals comes out of a team's attack weighed against its opponent's defence. A strong attack is damped when it meets a strong defence — which is why comparing the ratings on their own is misleading.
Attack: Paris FC score close to the league average.
Opponent's defence: Lyon's defence is close to the league average (0.92×), so it doesn't move the expectation much.
Result: These two effects give an expectation of 1.47 goals (including home advantage).
Attack: Lyon score close to the league average.
Opponent's defence: Paris FC's defence is close to the league average (0.95×), so it doesn't move the expectation much.
Result: These two effects give an expectation of 1.42 goals (including the away penalty).
A single scoreline is always less likely than an outcome: “home win” is the sum of 1-0, 2-0, 2-1 and dozens more. So the most likely single score can be a draw while the most likely outcome is a home win; the two do not contradict each other.
Ratings were computed from 1326 matches played in this league before kick-off; the team with fewer has 37 matches. No data from after the match was used. The model's track record is published unfiltered.