Skip to main content
Sensible Stats / Model Accountability

Locked before kickoff.
Never rewritten.

Every Champions League prediction is frozen to a public log hours before kickoff, with the measured lead time recorded per fixture. When the model is wrong, that stays on the record too.

Log updated 2 Aug 2026, 21:41 UTCModel v0.8.0← Champions League hub

Fixtures settled

0

0 predictions frozen & waiting

Outcome accuracy

0 correct calls

Exact scores

Predicted scoreline hit

Avg Brier score

Lower is better · 0.667 = guessing

The record

Frozen pre-kickoff probabilities against final results, newest first. Knockout legs are graded on the 90-minute scoreline; a tie decided in extra time or on penalties still records who went through.

The record opens with the 2026/27 league phase.

Predictions freeze automatically inside the 12-hour pre-kickoff window once fixtures are published in September. Nothing is ever back-filled: a match that finishes without a frozen prediction stays unscored.

How the log works

  • A daily job freezes every fixture inside the 12-hour pre-kickoff window and records the measured lead time — the number shown is the one written down, never a rounded promise.
  • Results are settled against the frozen probabilities — outcome call, exact-score check, and Brier score — with every change committed to a public git history.
  • Knockout legs are graded on the regulation scoreline: the model prices 90 minutes, so a shootout is never counted as a "predicted" win. Who actually advanced is recorded alongside, so two-leg calls can be scored too.
  • Later model improvements cannot touch past entries. If the model changes mid-season, the old predictions stand as made.
  • All percentages are model confidence from simulation — analysis, not betting advice.