Where the data comes from and how often it updates

Every figure on Tofiko comes from one of three external sources, or from a model built on top of them. This page says which is which, how fresh each one is, and why some matches carry a model probability while others only carry a price.

The sources#

7msport is the spine. Fixtures, kick-off times, results and the odds you see in the three price columns on a match row all come from here. When the site shows a home/draw/away price with a previous price beneath it and a movement arrow, that is 7msport data captured at two different moments.

API-Football powers the external signal layer — the extra data available on the MAX plan — and supplies World Cup fixtures and related tournament data.

Open-Meteo supplies weather for fixtures. Conditions matter more in some leagues and some months than others, which is exactly the kind of context a goals model can use.

How often things refresh#

• Odds refresh on a schedule through the day. The price on a row is a recent snapshot, not a live ticker. The previous price shown beneath it, with the ↑ or ↓ arrow, is what makes movement visible — that comparison is the point of storing both.
• Live scores update continuously while matches are in play. The Live view and the live minute on a match row track the game as it happens.
• Everything else follows the fixture. Results, form dots and standings settle after matches finish.

If a live score looks frozen or a match disappears from the list, that is a known behaviour rather than a display bug — feeds occasionally drop or revert, and we withdraw a live claim rather than show a stale one. Troubleshooting covers what to do.

The models#

Three models run over that data and are blended into a single ensemble:

• Poisson — models the number of goals each side scores as a rate, then turns those rates into probabilities for every scoreline. How the Poisson distribution works walks through the arithmetic.
• Dixon-Coles — a refinement of Poisson that corrects its known weakness at low scores, where plain Poisson misprices results like 0-0 and 1-0.
• Elo — ignores goals as such and tracks team strength as a rating that moves after each result.

The first two are goal-based and the third is strength-based, which is the reason for combining them: they fail in different ways, and a blend is steadier than any one of them alone. How our models work covers the reasoning in more depth.

Why coverage varies by league#

Odds coverage is wider than model coverage, and that gap is visible on the fixture list.

A model can only be fitted to a league with enough reliable history behind it. Without that, any probability it produced would be a guess dressed up as a number — so no probability is shown at all. Those rows appear as odds-only in the market chips legend, alongside model-backed and unavailable.

What this means when you are reading the list:

• Model-backed row — you have a price and an independent estimate, so a comparison is possible.
• Odds-only row — you have the price and its movement, and nothing to check it against. Market movement is still information.
• Unavailable — no market data for that selection.

Coverage widens as history accumulates, so a league that is odds-only today may not be permanently.