Track record:  3-month windows checked ·  inside the 90% band () → every one dated, right or wrong

Were we right?

Every forecast is saved with its date. When the date arrives, the real price goes next to what we said — right or wrong, nothing deleted. The tally first; the rules below it.

resolved 3-month windows checked
inside the 90% band
covered names, each with its own public record
0
forecasts deleted or rewritten. Ever.
MarketNamesWindowsInside the 90% band

Graded so far

NameHorizonSaid (p5–p95)Real closeResult

The live ledger — forecasts awaiting their date

showing

Published cones whose check date hasn’t arrived yet. Each will be graded on the real close of its date — projected dates use the exchange’s own calendar.

NameAnchoredAt priceHorizon90% band (p5–p95)Grades on
From 27 July 2026 every new forecast is checked at 1 month and 3 months from the day it was published — real calendar dates. If that date is a weekend or a public holiday on that exchange, we check on the next day it actually trades. Nothing else moves it. The engine’s backtest measures the same calendar window we publish, instead of a fixed session count.
Each forecast gives a likely range. When its date comes, we see whether the real price landed inside that range, and keep a running tally. If our "9 times out of 10" ranges are honest, the price should land inside them about 90% of the time. The big tally above is the backtest record across every covered name’s resolved, non-overlapping 3-month windows; each study page carries its own name-level record.