Note de recherche

Calibrating predictions differently for friendlies vs tournaments

Statut : Shipped (Variant 4 — per-tier Platt temperature scaling). Production calibrator uses a hybrid strategy: Platt for the tournament tier (where isotonic collapses to identity at n~70), isotonic for friendlies/qualifiers (where it's more expressive at n~400+). Gate passed. See results belowRésumé + note complète · 3,433 mots

Résumé

The shipping ensemble calibrator (scripts/fit_ensemble_calibrator.py) fits per-class isotonic regression curves on the uniform-averaged three-component output (Elo bracket MC + Dixon-Coles + Hierarchical Poisson MAP). The first cut lifted holdout ECE on the 365-day common-subset training pool from 4.62pp uncalibrated → 2.70pp under the pooled-across-tiers fit (5-fold CV, n_train = 939, current artefact at data/wc2026/ensemble_calibrator.json).

A subsequent tier-aware refit (three sets…

Note complète

Standard Pass

Lire la note de recherche complète

Calibrating predictions differently for friendlies vs tournaments fait 3,433 mots. Le Standard Pass débloque chaque note de recherche en intégralité, plus la prévision complète et les notes par équipe et par joueur, valable pour tout le tournoi.

Obtenir le Pass — 15 $

24h self-service refund·No subscription, no auto-renewal·Access through 31 Dec 2026. See refund policy.