Calibration scorecard.

Every forecast Phronesis has ever minted. Issue date, vertical, archetype, canonical quantiles, inference cost. Resolution status accrues as evaluation windows close. The track record is uncopiable; the substrate is publicly auditable.

Data fetched live from the Hermes contract surface at phronesis-jrstinehour.replit.app/scorecard.json. Cached at the browser for 5 minutes.

A track record that compounds.

Every forecast carries an issue date and an evaluation window. When the window closes — a quarterly disclosure, an annual report, an actual outcome — the ground-truth value is published next to the forecast on this page. Calibration accrues over time.

A forecast claiming a 70% probability that BTC will cross $X by Q3 isn't right or wrong on its own; it's part of a population. If we say 70% across 100 such forecasts and 70 of them resolve in the predicted direction, the substrate is calibrated. If 30 of them do, the substrate is overconfident. If 90 of them do, it's underconfident. Every customer can see this number for every vertical, every archetype, every issue period.

This is what verified-correctness means. Not an assertion in a marketing deck. A computable property of the published forecast ledger.