How the model
actually works.

Why are there no predictions right now?

Matchday's published predictions are paused across every competition while the forecasting model is rebuilt on a new data engine. The replacement engine doesn't yet hold data for every sport published here, and running predictions for some sports while guessing at the rest isn't a trade worth making. So all picks are on hold until that coverage is complete.

Everything that isn't a prediction keeps working normally — the pause is on the forecast, not on the site.

What still works while predictions are paused?

Nearly all of it: fixtures and results, live and final scores, standings and brackets, team and player stats, Elo and opponent-adjusted ratings, market odds wherever a provider supplies them, news, and the complete graded history of every pick made before the pause.

The only thing withheld is a new pick, confidence figure, or edge on a game that hasn't been played yet.

Why pause instead of just switching models?

Because the replacement isn't ready to publish, and saying so beats shipping it anyway. Three things have to be true first: it needs data for every competition Matchday covers, its own pipeline has to promote its forecasts out of shadow status, and it needs a graded sample large enough to support a calibration claim rather than a hunch.

None of the three is met today. A model that hasn't cleared its own bar doesn't get to borrow this site's.

When do predictions come back?

When the data gaps are closed and the rebuilt model has cleared that gate on measured results. There's no date attached to it, and inventing one here would be its own kind of dishonesty. Publication resumes sport by sport as each clears — not all at once on an announcement.

Does the pause change the existing track record?

No. Every pick published before the pause stays exactly as it was and continues to be graded against official final results. Nothing is rewritten, withdrawn, or quietly re-scored, and no pick is added after the fact for a game played during the pause.

The scorecard stays the real record of what the model did while it was publishing — which is the whole point of publishing before kickoff in the first place.

Who builds and runs Matchday?

Matchday is a passion project, currently built and run by one person rather than a company or a team. It's kept intentionally small, free, and focused on getting the model and the data right. If it ever grew a lot, that would mean rethinking how it's maintained and funded — that's a problem for if and when it happens, not something the product is built around today.

How can I use Matchday?

However works for you — settling an argument, following the model's own track record, comparing it with the market, or using the information in your own decisions. Matchday isn't paid by anyone with a stake in what you decide, and it isn't affiliated with any bookmaker.

How does Matchday calculate a win probability?

It blends several factors depending on the sport: season record and points, recent form, scoring margin, home advantage, rest days, an Elo rating, an opponent-adjusted rating similar to strength of schedule, and where available a team class or poll rank. Each factor's pull toward one side is shown transparently in the match breakdown.

What does confidence mean?

Confidence is the model's estimated probability that its picked side wins, shown as a percentage. It is an estimate, not a guarantee — games with limited current-season evidence are intentionally shown with more conservative confidence.

What does edge mean?

Edge is the gap between the model's probability for an outcome and the betting market's implied probability for the same outcome. It highlights where the model and the market disagree most so you can decide how that comparison matters to you.

What is Elo rating?

Elo is a rating that rises after wins and falls after losses, weighted by the strength of the opponent — the same style of rating used in chess, adapted per sport and updated automatically after every graded result.

What is the opponent-adjusted rating?

It is a rating built only from a team's own results and the strength of who they played, correcting a team's raw scoring margin for whether their schedule has been easy or hard so far.

Why does the model sometimes disagree with the betting market?

The market can price in information the model does not see yet, like late injury news or sharp betting activity. The model can weigh things the market is slower to move on, like recent underlying performance. Disagreement is reported as edge, not treated as the market being wrong.

What does Upset Radar mean, and who is the underdog?

The standings and betting market are both required. The candidate must be clearly weaker in the current standings and must also have the lower consensus market-implied win probability. Reputation alone and Matchday's original favorite do not define an upset.

Upset Radar requires two conditions at once. First, the candidate must be clearly weaker in the current standings, based on season win rate or points per match with enough games played. Second, that same team must be a genuine market underdog and Matchday must give it a win probability at least 8 percentage points higher than the market does. For example, if a lower-standing team is priced at 22% by the market and 31% by Matchday, the radar sees a +9-point underdog edge.

A radar flag is not automatically the official pick. It says the underdog looks more competitive than the market price suggests. Actually changing the official pick to that underdog requires stricter checks: the adjusted probabilities must be close, the matchup must have a high upset/volatility score, and the market gap usually cannot be too wide. If current market odds are unavailable, Matchday normally will not make a pregame upset pick.

How does Matchday track its own accuracy?

Every pick becomes eligible for a verified public lock inside 12 hours of kickoff and is graded only after an official final result; picks are never rewritten afterward. The pipeline verifies that both the lock and grade persisted before a fetch succeeds. In knockout games, the model pick is graded by the team that advances after extra time or penalties while the displayed score remains unchanged. Betting-market comparisons follow the market's own settlement rules. The Scorecard tab shows the full hit rate, calibration by confidence band, and which factors have actually been predictive recently.

Why didn't a pick change even though the market moved?

Picks become eligible for a verified lock inside 12 hours of scheduled kickoff and stay locked. If odds arrive later, only the market-comparison fields around the prediction can be filled; the selected side and confidence cannot change. That's deliberate: a model that could quietly rewrite its own picks after seeing new information wouldn't have an honest track record.

Where does Matchday's data come from?

Licensed and free-tier sports data providers per sport, plus a market-odds provider and public RSS news feeds. The full list of sources and their terms is on the Legal page.

Do I need an account, and is my data private?

No account is required. Personal settings like favorite team, watchlist, and appearance stay in your own browser's local storage rather than on a server. Full details are on the Legal page's privacy section.

Is Matchday a live-score service?

No. Matchday is a pregame and postgame analysis product. In-progress games remain Result pending until the hourly pipeline receives a provider-confirmed final result suitable for grading. It does not promise live clocks, score alerts, or minute-by-minute updates.

How often does the data update?

Production runs an hourly pregame and postgame fetch. Odds are requested only for upcoming games close to kickoff and cached to conserve quota. Browser reloads read the latest published dataset; they do not force providers to publish a result sooner.