Dynamic service

Submit results

Submission is not openThe steps below describe how it will work. Contributing tasks, checks or grading references is a different route and is open by e-mail: see Contribute.

Register the harness (and its tools) and provide a callable model × harness.
The site runs it (site-run), or re-grades a run returned on a task package (submitted-run re-grading), and generates the result record and run records.
The record enters the registry by pull request, validated against result.schema.json and cross-referenced to the benchmark version, model and harness.
Site review: reference solution full score, env_delta, run records verifiable, trajectory sample checked, canary check. Only then is verified_by_site set and the row listed.