Real courts
Visible and active games, camera conditions, contamination, omissions, corrections, and failures.
KhelSutra Evidence
An open framework for comparing sports-video systems without turning a model score into a coaching claim.
Checked-in examples are synthetic contract fixtures—not product evidence or a blind benchmark. What is published, and what is not, is listed in the release index below and is rendered from the registries rather than asserted here.
Both registries are empty, so no release is published here yet. This sentence is rendered from the registry data, not asserted alongside it.
Published and signed off through the KhelSutra evidence programme.
No official releases.
Self-run evidence that passed contract and safety review. Not KhelSutra verification, endorsement, licence clearance, or a quality award.
No community releases.
Release data digest 3ad3988894d88cf246010023af38c85a075d5a053f75ce48768e7cabac7f537f — the SHA-256 of the canonical projection of the registry entries above.
Visible and active games, camera conditions, contamination, omissions, corrections, and failures.
Licensing across code, models, data, media, APIs, codecs, and outputs—unknown is never green.
USD and accelerator seconds per source hour, retries, manual minutes, and accepted-output cost.
Held-out improvement, export, clean restore, prediction replay, promotion, and rollback.
Confirmed facts, model observations, calculations, coach assertions, and unknowns remain visibly distinct.
Validate, score, compare, and package your own work locally; receive a rationale rather than a leaderboard badge.
KhelSutra is an AI agent for the coach. It is never the AI coach. It retrieves, measures, clips, assembles, and explains. The coach—or a person coaching themselves—interprets the evidence and decides what happens next.