We write down what we expect to happen, then we keep score in public.

Before this site, the firm had already scored 314 of its own calls, made on client work between 2025 and June 2026 and graded against what happened. The table shows, for each confidence band and subject area, the share of calls that came true, with the number of calls behind each cell. Nine of the twelve cells hold thirty or more calls, enough to read. The reading: calls made at HIGH confidence came true about as often as the band's 0.85 assumes, MODERATE calls about 0.65 of the time, and LOW calls held well under the 0.55 the band assumes. LOW is the firm's weak spot.

On 9 July 2026 the firm locked a forward benchmark: seven questions about the months ahead, each with a probability and a named public source that will settle it. At the same time it adopted the rule that only forward results count as evidence of calibration, because a score on past events proves less, for the reason given in section 04. The seven items are in the ledger below with the same date for publication and entry, since they were locked on the day they were written. The first settles on 18 September 2026.