Your next match.
A clearer perspective.

A transparent way to read a match.

1. Choose your competition and dates

Browse every match returned for the selected date range, with search, status filters and pagination. “All leagues” lists the matches your connected provider can return for one UTC date. API plans may restrict competitions, seasons, forecasts and update speed.

2. Understand the source

API-Football supplies primary three-way forecasts. When those are unavailable, football-data.org can supply match history for JK’s experimental goal model. Fallback is possible only where the league and the same match can be verified. We never silently substitute a different game.

3. Read the probabilities

Home, draw and away total 100%. The predicted outcome is the largest probability; a draw can be the leading outcome. A favorite can still be more likely to draw or lose than win. A close forecast may produce a recommendation to skip the match.

4. Treat tips as probability-based suggestions

Double chance combines two mutually exclusive match outcomes. Goals and both-teams-to-score suggestions appear only when the goal model supplies estimates. When the market consensus is the source, probabilities come from bookmaker prices with the margin removed, so they carry no edge over the bookmakers; otherwise no bookmaker prices are used. Either way we cannot claim profitable odds or value. All markets refer to 90 minutes plus stoppage time.

Experimental model limitations

The fallback rates every team on attack and defence using recent regulation-time results from football-data.org, including each team’s other competitions. Newer matches count more, ratings are shrunk toward the average, and the opponent’s strength is taken into account. Home, draw and away chances come from independent Poisson goal distributions. It was tested on about 1,800 league matches from 2025/26 across the five major leagues (each predicted only from earlier results): it clearly beats naive baselines and the older model, but is a little less accurate than bookmaker closing prices. It has not been tested on Champions League fixtures, ignores injuries, lineups and rotation, and can compare teams from different leagues only roughly. When a team has too little history, an older, simpler goal-average model is used, and when even that is not possible there is no forecast.

Site setup check

For whoever runs this site. Enter the CRON_SECRET you set in Vercel to see what is configured and what needs fixing. It is sent only to this site, over HTTPS, and is not stored. Nothing about the setup is shown without it.

Pick records

Each pick preserves its original forecast. You can remove a pick until kickoff; after kickoff, or once it has a result, it is locked and cannot be changed or removed, so results cannot be hidden. Results are checked on request, three eligible matches at a time. Delayed, postponed or incomplete results stay pending. Extra-time and penalty outcomes are not substituted for regulation-time scores. Browser profiles are separate from the earlier ChatGPT-hosted app’s accounts and picks.