La Liga matchday 1: what a round played across twelve days does to a model
Published
What a postponed matchday does to a prediction model is expose something the fixture list normally hides: “matchday 1” is an administrative label, not a moment in time. La Liga’s opening round of 2026-27 began on 15 August and did not finish until 27 August. Twelve days, two other rounds played in between, and by the time the last matchday-1 fixture kicked off, several of the teams involved had already played twice more.
That is not a scheduling curiosity. It changes what a results-driven model knew when it priced each game, and it makes the round genuinely ungradeable as a round. The interesting thing about La Liga matchday 1 is not who won. It is that ten fixtures carrying the same label reached us in ten different states of knowledge — and that three of them reached us too early to be forecast at all.
Why the round was stretched
The cause was the 2026 World Cup. Yahoo Sports reported that Barcelona’s opener against Athletic Club was postponed because of Jules Koundé’s continued involvement in the tournament after France booked a semi-final place, and a further Yahoo round-up confirmed that matchdays 1, 2 and 3 all had games moved. beIN Sports covered Barcelona vs Athletic Club as a live matchday-1 fixture on 27 August — the round’s final game, played nearly a fortnight after its first.
Three fixtures that are not on this site at all
The honest place to start is with what we do not have. Our model has no stored prediction for Alavés vs Getafe, Sevilla vs Rayo Vallecano or Racing Santander vs Villarreal. Those three were played on 15 and 16 August, before our first La Liga model run produced anything.
The consequence is stricter than it might sound. A match page on this site is built only where a prediction exists, so those three fixtures have no page here — no probabilities, no result card, nothing. They are absent rather than unscored. That is deliberate: back-filling a probability after a result is known is the one thing a public track record cannot survive, and a page showing a result with no forecast beside it would invite exactly that comparison in a reader’s head.
So a quarter of the round sits outside our record entirely. Saying so is cheaper than the alternative, which is quietly letting the seven fixtures we did price stand in for ten.
The same label, very different evidence
Of the seven we did price, look at the two ends of the range.
Espanyol vs Levante was forecast in our 16 August run at 46.1% Espanyol, 25.0% draw, 28.9% Levante. At that point the model had seen no 2026-27 La Liga football whatsoever. Every number in it came from prior-season ratings and whatever regularisation pulls a new season’s estimates toward the middle. Espanyol won 3-0.
Barcelona vs Athletic Club was forecast in our 26 August run at 73.3% Barcelona, 15.8% draw, 10.9% Athletic. By then the model had two full rounds of current-season results, including Barcelona’s own. Barcelona won 2-0.
Both are matchday 1. One is a nearly evidence-free prior; the other is a genuine in-season forecast. A reader who lines those two numbers up as though they were produced by the same process is comparing a guess to an estimate. The confidence gap between 46.1% and 73.3% is partly about the teams and partly about the calendar, and there is no way to separate the two from the outside.
The result that went furthest from the model, and why
Celta Vigo vs Osasuna is the round’s clearest miss. Our 26 August run made Celta the favourite at 50.3%, with Osasuna at 24.5% — a shade over two-to-one against the away win. Osasuna won 2-1 in Vigo.
The match report is worth reading before deciding what that says about the model. Per ESPN’s match data, Iago Aspas put Celta ahead in the 14th minute; Marcos Alonso was sent off in the 51st; Enrique Barja equalised in the 54th and Ante Budimir scored twice, including a penalty, to finish it. Celta led, and then played roughly forty minutes a man short.
Our pre-match number contains no information about a 51st-minute red card, because nothing can. This is not a defence of the 50.3% — a model that priced Celta’s win at half and watched them lose is a model that was wrong on the day, and the track record counts it as exactly that. It is a caution about the inference. “The model overrated Celta” and “the model could not see a sending-off” produce the same scoreline and call for completely different responses, and only one of them is a reason to change anything.
Two other home favourites also failed to win. Deportivo vs Elche was priced at 49.8% for the home side on 16 August and finished 1-1. Valencia vs Real Betis was our least confident call of the round — 39.9% Valencia, 33.9% Betis, a near-coin-flip — and Betis won 1-0. That last one is worth noting precisely because it was not a confident call: a 39.9% favourite losing is the ordinary outcome of a fixture the model said it could barely separate.
Where the calendar helped rather than hurt
The round’s two most confident predictions were also its two latest, which is not a coincidence.
Real Madrid vs Real Sociedad was priced on the morning of the game at 67.0% Real Madrid. It finished 4-1. Al Jazeera’s live coverage records a Kylian Mbappé hat-trick with Vinícius Júnior also scoring, and notes that Luka Sučić had made it 1-1 at the interval — the game was level at half-time before it was not.
Atlético Madrid vs Málaga, priced at 58.3% on 16 August, finished 2-0. That one sits in the middle of the range in both senses: a mid-August forecast on thin evidence, and a mid-range probability.
The pattern to take from this is not “the model gets better later”. It is narrower and more mechanical: a round spread over twelve days gives the fixtures at its end an information advantage that the fixtures at its start could not have had, and any reading of “how matchday 1 went” that ignores that is reading a scoreboard, not a forecast.
What this means for matchday 3
La Liga’s third round starts on 29 August with the table still distorted — teams have played different numbers of games, and some of the postponed fixtures from matchdays 2 and 3 are still to be rescheduled. Our model handles that the same way it handles everything else: it uses the results that exist and does not impute the ones that do not. That means the next few rounds of La Liga probabilities are built on unequal samples between clubs, and they will stay that way until the calendar catches up.
The per-match numbers, with the model run that produced each one, are on the seven match pages linked above. The performance record is on the track record page, which is the only surface where we publish it.
The matches, at a glance
Primera DivisionFT
Espanyol3
Levante0Primera DivisionFT
Deportivo1
Elche1Primera DivisionFT
Atlético Madrid2
Málaga0Primera DivisionFT
Valencia0
Real Betis1Primera DivisionFT
Real Madrid4
Real Sociedad1Primera DivisionFT
Celta1
Osasuna2Primera DivisionFT
Barça2
Athletic0