Portugal 2–1 Croatia: how the Jarvis supercomputer's prediction held up

How the Match Unfolded

The match between Portugal and Croatia ended in a 2–1 win for Portugal. Three goals were scored in total, no small number set against the backdrop of the pre-match forecast. The route there was less smooth than the scoreline suggests: Croatia actually went ahead through Ivan Perišić on 53 minutes. Cristiano Ronaldo levelled from the penalty spot on 68, and it took until the fourth minute of stoppage time for Gonçalo Ramos to score the winner. So Portugal won by a single goal over Croatia, exactly the direction the Jarvis supercomputer model had leaned toward before kick-off.

Match Highlights

The first point worth noting is the final result itself: Portugal, the side seen as stronger going in, held onto the win. Second, the match produced three goals, a number that points to an open, back-and-forth contest rather than a stalemate. Third, Croatia weren't shut out: the losing side still got one on the scoreboard and kept the final gap to a single goal. Put those three details together and you get a familiar picture for matches with a clear but not overwhelming favourite - the stronger side wins, while the other still manages to leave its mark on the scoreline.

Jarvis Checked Against the Result: Pre-Match Forecast vs. What Actually Happened

This is the heart of the piece: setting what Jarvis produced before the match next to what actually happened. On the winner, the model rated Portugal the clear favourite at 55%, against 27% for a draw and 18% for a Croatia win. The final result, a 2–1 win for Portugal, landed squarely in the scenario Jarvis rated most likely of the three before kick-off.

On the scoreline, the model put forward a predicted score of 2–0. It's worth being precise here: this was never meant as the single most likely scoreline, but a number Jarvis uses to sum up the match's overall lean. The actual result of 2–1 missed that predicted score by exactly one goal. That gap lines up with a detail already built into the pre-match data: the forecast's confidence level was rated LOW. When confidence is tagged low, the possible gap between the predicted score and the real result widens, and a one-goal difference sits well within that range - meaning the model had already allowed for this margin, rather than simply getting it wrong.

On total goals, Jarvis projected an expected 2.5952, alongside a 48% chance of the match producing three goals or more. The three goals actually scored sit right at the edge of that expected range, if anything slightly above it. Taken together across all three layers of data - the winner, the goal-scoring trend, and the predicted scoreline - the model got the first two right and missed by just one goal on the specific score.

What the Expected-Goals Number Actually Tells You

The number in Jarvis's output most likely to be misread is the expected goal total of 2.5952. This isn't a claim that the match would produce roughly two and a half goals, since no match can have half a goal. It's an average pulled from many different scoreline scenarios the model weighs, each carrying its own probability weight. Some scenarios are low-scoring, some are high-scoring, and 2.5952 is the balance point across all of them.

Alongside it sits a more concrete figure: a 48% chance of the match reaching three goals or more, meaning nearly half of the scenarios the model built land in the high-scoring bracket. The actual match produced exactly three goals, sitting right in that upper group. That's the right way to read an expected-goals figure: it doesn't point to one single scoreline, it describes an overall trend - this match leaned toward being open rather than a stalemate, and the three goals scored followed exactly that direction. When you want to know whether a model is leaning toward a low-scoring or high-scoring match, the expected total together with the percentage at the three-goal threshold tells you far more than any single predicted scoreline can.

What the Probability Numbers Really Mean

One thing worth repeating every time you read numbers like these: the 55% given to Portugal was never a guarantee of victory, it's a relative level of confidence, set alongside 27% for a draw and 18% for a Croatia win. All three scenarios carry a nonzero probability, meaning all three were genuinely possible in reality, differing only in how likely the model considered each one.

Even a figure like 18% carries real weight rather than being something to wave off: statistically, that's a scenario expected to show up roughly once in every five to six repeats of a similar situation. In this match, that scenario didn't come to pass since Portugal still won, but the 18% is a reminder that the side rated as the underdog could well have produced a different outcome. That's also how a probability forecast should be read: it tells you the lean and how strong that lean is, not a single outcome locked in stone.

Data Source

Every forecast figure in this piece comes from the Jarvis supercomputer model and was produced before the match was played: the 55% probability of a Portugal win, 27% for a draw, 18% for a Croatia win, the predicted scoreline of 2–0, the expected goal total of 2.5952, the 48% chance of three or more goals and the LOW confidence rating. The scorers and goal times (Perišić 53', Ronaldo 68' pen, Gonçalo Ramos 90+4') come from the match record. This is statistical analysis, not betting advice.