Calibration

The record, whichever way it went.

1 graded. Too few to mean much. We will keep saying that until 100. Every Probability the desk publishes is graded after the series and kept here with its date.

Probabilities published: 7
graded: 1
hit rate: 0.0%
Brier score: 0.336

The model, in full — rank-and-flags, 28 Aug 2026

01Strength from rank. A team's Valve Regional Standings rank becomes strength = 1000 − 120 × ln(rank). The log is there because the gap between 1st and 5th is wider than between 30th and 34th.
02Base probability. The two strengths go through the logistic p = 1 / (1 + 10 ^ ((opponent − team) / 400)).
03Adjustments. Whole percentage points, each with a sourced reason, each printed on the Dossier. There are no unlisted adjustments.
04Clamp to 5–95 and round to a whole percent. Then it is graded, right or wrong.

The model is simple on purpose. It is published so the number can be checked, and it changes only in public, with a date.

Published Probabilities

series

the call

result

grade

A Dossier names one team in its Probability, so a 42% on that team is a 58% call for the other one. The column above states the call that way and grades it that way. The Brier score uses the Probability exactly as it was published.