Methodology
We do not try to prove success after the fact.
We seal what we believed in advance, then measure performance on those same immutable predictions.
Canonical forecast
Unique forecast ID, model/build/dataset fingerprint, track status and revision.
Sealing
A record sealed before kickoff. The sealed version cannot be overwritten.
Bitcoin timestamp
OpenTimestamps proof and online block-header verification. The proof file is not automatically published.
Proper scoring
Brier, Log Loss, calibration, MAE/RMSE and target-specific grading.
Data protection and publication
The frontend only reads publication-safe records. Raw features, training data, model binaries, full proof bundles and internal debug context are not exposed in public responses.
Detailed Menta data is restricted to approved accounts protected by mandatory 2FA.