Receipts

Reading a Brier: it squares the gap between the forecast logged with each call and what happened — 0 is perfect, 1 is certain and wrong. Each score sits beside what saying “50/50” to the same calls would have scored. That bar is lower than the textbook 0.25, because a partial counts as half and 50/50 gets a partial exactly right. Below 10 scored calls a score is marked provisional: shown from the first receipt so nobody has to take our word for when it started, but not yet a calibration.

Calls logged until Sep 6 were scored on the wrong number. The ledger wrote each call’s framework fit — how strongly our frameworks line up behind it, 0 to 100 — divided by 100, as if it were the chance of the call coming good: fit 99 was logged as a 0.99 forecast. Fit was never tested as a probability, and when we measured it in September, higher fit did not mean better odds. Those calls are scored exactly as logged, on their own — nothing is re-scored — and 49 more of them come due through Mar 2, 2027.

Calls logged since then carry a real forecast: the share of that signal’s resolved calls that came good, counting a partial as half, pulled toward 50% while the record is short. They are scored on their own.

A call is scored at the price when the daily scoring run reaches it, on or after its due date — not at a closing price. Where a stock split in between, the price is restated into the shares the call was made in, and the row says so.