Organizing a large corpus of ball trajectories into summarized clusters (mean point, direction vector, event count) so that similarity search against millions of historical events returns the top matching patterns in real time. Enables live game pattern matching against a full multi-season database rather than offline-only analysis. Uses cache management (most-recently-used eviction) for live contexts.
Pre-cluster all historical trajectories using DTW. For each cluster, store a summary: mean point, direction vector, event count, normalized composite distance. At query time, compare the live trajectory summary against cluster summaries (fast) rather than against all individual trajectories (slow). Return the closest matching historical patterns.