Lacrosse AI Video Analyzer — Development Brief

We are building an extremely accurate AI-powered lacrosse video analysis platform. The ultimate goal is for a user to upload or link virtually any lacrosse video—YouTube, MP4, cell-phone footage, Hudl/Veo-style film, individual clips, or full games—and have the system automatically understand and analyze the game. The system must identify and track the ball, teams, players/jersey numbers, game clock, possessions and field locations, then automatically divide the game into possessions and timestamp/tag events including faceoffs, passes, catches, ground balls, dodges, shots, shots on goal, goals, saves, turnovers, caused turnovers, penalties, clears, rides, substitutions and transitions. The objective layer is only the beginning. The system must eventually understand lacrosse tactics and decision-making: offensive formations and concepts, defensive structures, slides/recoveries, two-man games, cuts, drifts, replacements, spacing, leverage, passing/skip lanes, shot opportunities, defensive rotations, matchups and transition. The analyzer should determine not only what happened, but what should have happened and why. For example: Was a skip lane open? Did the player recognize it? Was the read correct but late? Did he miss a shooting opportunity? Did his movement create pressure or allow the defense to rest? Who actually created the advantage that produced a goal? Goalie analysis must go well beyond save percentage, including shot location/type, shot-on-goal %, positioning, angle, screens, reaction/step technique, save difficulty, rebound control, second-chance opportunities, communication, outlet decisions, passing accuracy, clearing and transition creation. Outputs should ultimately include full-game timelines, advanced statistics, player evaluations, development reports, team reports, shot/save maps, goalie reports, recruiting analysis, opponent scouting reports and searchable video clips tied to every event. Accuracy is the highest priority. Architecture should separate (1) objective video observations, (2) events/statistics, (3) tactical inference, and (4) coaching/decision analysis. The system must attach confidence/evidence to uncertain conclusions and return UNKNOWN/NEEDS REVIEW rather than hallucinating. We are creating a detailed Lacrosse Intelligence Standard (LIS) defining possessions, events, edge cases and expert coaching judgments. The development approach should be iterative: define → annotate real film → train/detect → compare against human ground truth → identify failures → add rules → retrain/test. Ultimate UX: Upload a lacrosse game → click ANALYZE → receive an accurate, searchable, possession-by-possession breakdown of every player and team, advanced statistics, tactical/coaching analysis, player-development insights and opponent scouting intelligence—with every conclusion linked back to the supporting video.