Machine-generated probability forecasts on questions drawn from live public signals — every forecast logged before its outcome is knowable, every outcome scored, every calibration result published, including the failures.
Live forecasts in the order they were issued. A forecast is never edited or deleted after issue — resolved questions show the real outcome next to what we said.
We hold ourselves to a statistical launch gate — sample size, skill over a naive baseline, and two independent bias tests — and publish every evaluation verbatim. Credibility is the audit trail, not the absence of failures.