Rounds

Evaluation rounds

An evaluation round runs a defined set of benchmark versions against enrolled models and harnesses under fixed conditions, and produces an auditable report. MedGauge provides the process and the records; it does not replace any official testing or certification.

Not open. No round is accepting applications yet. Partner institutions are named only with their written consent.

Kinds of round

Open evaluation rounds

Run by the site on a published schedule. Any registered model × harness can enrol. Results enter the leaderboard after review.

Partner rounds

Run with an evaluation institution on an agreed scope. Report format and publication are agreed in advance; results enter the public leaderboard only with the enrolled team's consent.

What every round page will state

Scope: task families and capabilities covered.
Fixed conditions (seven): model and version, service snapshot, reasoning setting, runtime environment, tool permissions, harness, budget.
Enrolment requirements and run mode (site-run, or submitted-run re-grading).
Timeline.
Report format: model × family table with the share of tasks passed in all three runs, per-family breakdown and failure attribution. Models not yet tested are shown as "not yet tested", never as zero.
Closed management and training isolation: task material is used only for evaluation and diagnosis and never for training; exposure retires the task.

Flow

Application → screening → review → approved → enrolled → testing → reported → closed. Every state change is logged; rejections carry reasons. See Apply (not open).