CITADEL VAR™

AI referee assistant · Phases 1–10

The engine recommends. The referee decides. Every recommendation and every human decision is written to a tamper-evident audit chain. Player positions shown in the review station are produced by the live vision pipeline, with the calibration error and position uncertainty shown alongside them.

checking API… health JSON

VAR Review Station

Calibrated offside line drawn from the tracked frame, frame-accurate transport, slow motion, measurement tools, natural-language search.

Referee Station

Incident queue, AI recommendation with its reasoning, and the decision form with a mandatory override rationale.

Fourth Official

Match control, discipline watch and live evidence-chain integrity.

Administration

Venues, camera angles, officials, matches and service health.

Tournament

Competitions, live standings and fixtures.

Analytics

Referee performance, pitch control and physical output — with every figure labelled as measured or withheld.

Vision pipeline

numpy-only: homography calibration, Kalman tracking, ball ballistics. Run npm run vision:demo to push a tracked offside through the pipeline into the review station.

Footage labeller

Click markers, players and the ball on real match video, record the referee’s answer, and export the files the real-footage harness scores.

To-Do list

Everything that needs budget (camera, mini-PC…) or a human to finish and sell VAR, with your own cost estimates.

Open To-Do list