Every project here makes its predictions in public, stores them before the event, and grades them after — nothing is edited after the fact. The point is to show whether a method works, honestly, over time.
A prediction only counts if it was stored before the thing it predicts. Each project writes its picks to a database ahead of time, with the reasoning, and grades them afterwards against what actually happened. Nothing is edited after the fact, and the misses stay on the page next to the hits.
A record means little on its own, so every project shows a public yardstick beside it. For sports that is the closing betting line — the best public forecast there is — and we say plainly when the model trails it, which for NFL games is the normal state of affairs.
The models are ordinary statistics you can read about on each project's own "How this works" page. Where a language model is used at all (for example, reading news reports for the NFL project), its input is limited, capped, logged with its source, and graded separately, so you can see exactly what it changed and whether that helped.
Nothing on this site is betting, financial or legal advice. The numbers are published to show whether a method works, and to keep us honest about it.
Each project lists its own sources. Sports data comes from the open-source nflverse project (CC-BY-4.0) and public market APIs; nothing paid, nothing scraped from behind a login. niftiIQ Lab is a project of Nifti Mobility Group.