The Method
A Retro-Prescient Audit examines a prescience claim the way an auditor examines a charter: it sets the claim against the record and scores the distance. It runs both directions in time.
Retro — take a foresight record someone claims and test it against the archive, misses counted. Prescient — seal new predictions under commitment before the outcome, so the forward record accrues where no one, the forecaster included, can quietly edit it.
The full definition is sealed in commit history: RETRO_PRESCIENT_AUDIT.md
- Pre-registration. A prediction counts only if written, with its failure condition, before the outcome.
- The keyed/keyless split. A hit deducible from priors already in hand is inference, not foresight — and is claimed only as inference.
- Misses print. Permanently. A run with no recorded misses is presumed corrupt.
- Named baselines. A hit counts only above what chance or base rates produce from the same inputs.
- Blind adjudication. The adjudicator sees neither forecaster nor stated probability, must cite evidence, and may return ambiguous.
The Instruments
Standing
This is an instrument being tested in public, not a validated one. Under thirty resolved predictions the score is noise — and the ledger says so on its own face. The record currently shows the machine missing more than it hits.
That number is the credential. A desk that prints its own bad news is the only kind whose good news means anything.
Consistency is not correctness — resolution is.