Audit the evidence
Review the frozen evaluation window, quantile coverage, pinball skill and the gates that kept the latest candidate out of production.
Open evidenceOpen project / Public verificationDon't trust the badge.
Don't trust the badge.
Inspect the boundary.
GERT is designed as public, falsifiable research software. Its strongest claim is not that every candidate works—it is that evidence, limitations and promotion decisions stay inspectable.
Inspect sourceReproduce the work
Inspect the public architecture, tests, model contract and data-provenance boundaries. Clone the repository and rerun the documented checks.
View repositoryChallenge an assumption
Report a reproducibility problem, propose a calibration test or identify an operational edge case through a public issue.
Open an issueCurrent authority state
The latest model candidate is not production-authorized.
It showed positive pinball skill but missed predeclared calibration tolerances. Live operating context remains useful; probabilistic prediction stays gated until a frozen candidate passes every promotion requirement.
Contribution standard
Useful criticism beats passive attention.
Reproducibility reports, calibration alternatives, operational edge cases and documented failures are the highest-value contributions to this project.