In the tipster rating, next to people, there are authors marked “model”. They are not a marketing showcase of “AI predictions” but participants under the common rules: they publish picks before kick-off, their odds are captured by the server, and their results are settled from the match outcome. Below is what is known about them and what is not.
Who is who
There are several models, and each is responsible only for its own markets. Overlaps are not a dispute but a comparison: on the same match two models may rate the goals total differently, and both bets go into the record.
- TheAnalyticD model — match result, goals totals and both-teams-to-score across all leagues.
- TheAnalyticX model and TheAnalyticF model — match result; two independent estimates of the same market.
- TheAnalyticT model — goals totals.
- TheAnalyticC model — corner totals.
- TheAnalyticK model — yellow card totals.
- TheAnalytic model — the rating model: corners match result and fouls totals.
The market name always stands next to the pick: “1” on corners and “1” on goals are different picks, and the match pages label them.
Rules common to all models
- A model publishes a pick only when its probability estimate for the outcome is higher than what the odds imply. No disagreement with the line, no pick; on some days a model has zero publications, and that is normal.
- One bet per match, a fixed share of the bankroll. The stake is not “tuned” to confidence.
- Publication goes through the same mechanism as for people: before kick-off, with the reference line’s odds locked, with no right to edit. How it works is in the article on locking picks.
- Models enter the rating on the same terms, with the same minimum samples as people.
Archived picks
For some models the history starts before the verifier went live: their estimates at the time are confirmed by timestamps, and the odds are taken from the opening line. Such picks are marked as archived, shown in the profile with a caveat, and do not take part in the rating. A model’s live history is only what was published in real time.
How to check a model
The same way as a person. A model’s page has everything any author has: settled picks, ROI and the bankroll curve, CLV by market and league, a confidence interval. The agreement column in the match table is separately useful: it shows where the model diverges from the majority of authors, and it is on those matches that you can see who adds value — see the article on consensus.
What the service does not disclose
How the models are built — features, training data, calibration method — is not published and will not be: that is what makes their results worth something. What is disclosed is exactly what is needed for verification: every pick, its odds at publication, the closing price and the outcome. A model cannot be judged by its description; it can be judged by its track record, and that is fully open.
What the models are not
A model’s pick is not a recommendation by the service and not a “signal”. TheAnalytic does not accept bets; the models take part in the verifier on the same terms as people so that readers can compare the two on identical, verifiable numbers.