Football Proof AIProbabilities with a public recordEvidence library

Free · Browser-only · Reproducible

Football prediction evaluation tools

Move from a single probability to a complete model audit. Each tool fixes its input contract, formula version, exclusions and evidence boundary before it displays a result.

Published by Football Proof AI · Updated · Educational analysis, not betting advice

Evaluation stack

Six questions, one evidence-first workflow

Start with the question your data can actually answer. A larger dashboard does not repair missing timestamps, mismatched matches or a sample selected after the result was known.

One forecast history

Accuracy Audit Lab

Audit a complete prediction CSV for timing, exclusions, hit rate, uncertainty, Brier, log loss and calibration.

Open tool
Two paired models

Model Comparison Lab

Compare model A with baseline B on identical settled matches using paired probability losses and uncertainty.

Open tool
Probability quality

Brier Decomposition Lab

Separate multiclass Brier error into reliability, resolution and uncertainty estimates, then compare honest baselines.

Open tool
Validation integrity

Temporal Leakage Auditor

Inspect walk-forward folds for post-kick-off predictions, unavailable features, train/test overlap and other temporal leakage.

Open tool

What every result must retain

A number without its denominator is not evidence

  1. 01

    Input identity

    A SHA-256 fingerprint binds an audit receipt to the same local input bytes.

  2. 02

    Time integrity

    Pre-match tools reject or expose rows that use forecasts or features after their allowed cutoff.

  3. 03

    Metric convention

    Summed and class-averaged multiclass Brier scores are never silently mixed.

  4. 04

    Inference limits

    Approximate intervals and planning calculations state their independence, variance and sampling assumptions.