How it works
Your film. A fixed camera or a phone. No wearables.
Selke works from the film a program or a player already has, from a rink camera or a phone. Computer vision turns the footage into positions, event detection finds the breakouts, entries and turnovers, and the outcome model learns which positions led to which results on that team's own film. The grade comes from results, not a rulebook we wrote.
Four steps
- PositionsEvery skater and the puck, in ice coordinates, from the camera you have. Both teams, every frame.
- EventsExits, entries, turnovers and takeaways found in the film. These are the labels, and the labels are what make the grade honest.
- OutcomesThe model learns which positions led to clean exits and controlled entries, and which led to turnovers, inside this team's system.
- The gradeGhost gap times leverage, priced in expected goals, shown over the clip with the reason in plain language.
What we grade, and what we refuse to
Positional value, ice and lane coverage, support angles, anticipation. Those are impact. Distance skated, top speed and hit counts are activity, and Selke will not headline them. A player can skate the most miles on the ice and be in the wrong place for all of them.
The workflow, the grade, the ghost view and the report are built and running on a demo season. The vision engine that reads a program's own film is in build. We are taking a small number of founding programs, agencies and players for the first graded seasons.
Book a walkthrough.
Thirty minutes. Three reps graded in front of you on the demo season, the ghost toggled, the report in your hands. No deck.
Book a walkthrough