Research note

Hierarchical Poisson — full PyMC NUTS posterior

Status: Shipped as a methodology surface (posterior-median P(H/D/A) gates parity with MAP; per-fixture credible intervals now publishable)Backtest date: 27 May 2026Topline + full note · 1,696 words

Topline

The shipping Hierarchical-Poisson fit (scripts/fit_hierarchical_poisson.py) is a MAP point estimate via L-BFGS-B. It is fast and is the production prediction surface, but it gives no measure of uncertainty on per-fixture P(H/D/A). The improvements-roadmap (§1.3) flagged this as the next methodology upgrade — not a Brier-chasing change (point-estimate Brier is already at the calibration plateau) but a publishable methodology one: the model now reports how confident it is in each forecast, a…

Full note

Standard Pass

Read the full research note

Hierarchical Poisson — full PyMC NUTS posterior runs 1,696 words. The Standard Pass unlocks every research note in full, plus the complete forecast and per-team and per-player ratings — valid through the tournament.

Get the Pass — $15

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