MedGauge Methodology, version 0.1.0

The normative text of the MedGauge benchmark for medical AI agents: what is measured (ten dimensions, each reported on its own), how tasks are described (ten task labels), how runs are graded (versioned grading programs reading declared fields of a structured work product), and how results are stated (an exam outcome per benchmark version, with the one-sided 95% Clopper–Pearson lower bound that a run of passed tasks supports, and its conditions). It also sets the rules for task acceptance, contribution, versioning and correction. Task materials, answer keys and verifiers are closed; every rule that decides a grade is public. This record is the first public version. Later versions are deposited as new versions of the same record.