Methodology — and how to check us

The short version of how the numbers work, and the part that matters more: how to verify that a prediction we say we made before kickoff really was made before kickoff.

The model, in three sentences

Every team carries a dated Elo rating (eloratings.net) frozen before each tournament. Match forecasts map the Elo win-expectation 1/(1+10^−Δ/400) through a calibrated independent-Poisson scoreline model (2.5 total goals, neutral venue) — 90-minute W/D/L, likeliest scores, and advancement with draws resolved through extra-time/penalties in proportion to win-expectation. We tested whether squad market value beats this out of sample: it doesn't, which is why the model stays ratings-only. Full detail on /method.

The lock system

Before each round we freeze the forecast into a static file under /locks/. The file's SHA-256 is committed to the public git history and tweeted before kickoff, then every deploy re-verifies the served bytes against that hash and aborts on any mismatch. The numbers cannot move after the fact — that's the whole point.

Check it yourself, right now

In any terminal (macOS/Linux — on Windows use WSL or Git Bash):

The output must be exactly:

That hash was published before the 2026 final kicked off. If the command ever prints anything else, the file changed after publication — call us on it.

The Cliff Index formula

/cliff-index ranks every 2026 squad by 0.5·z(weighted age) + 0.3·(last-dance minutes share) + 0.2·(1 − debutant share) — z-scored within 2026; squad age weighted by World Cup minutes (basis and coverage stated on the page); “last dance” = aged ≥31 on 2030-06-01. The formula is fixed as written — v1, no tuning.

Every locked artifact

Same command works for each: curl -sL https://thecliff.football/locks/<file> | shasum -a 256.

full method & prediction ledger → · the model scorecard →