```json { "table_name": "2010_WEEK 14_plays", "primary_key": "Season,Week,AwayTeam,HomeTeam,DriveNumber,PlayNumberInDrive", "candidates": [ { "rank": 0, "partition_key": "TeamWithPossession", "partition_key_label": "E", "partition_key_reasoning": "RPUV=82.91 is ideal for partitioning. 32 unique teams with balanced distribution (top 3 coverage 11.50%). Semantically represents the entity (team) whose actions are being tracked in each play event.", "clustering_key": "PlayNumberInDrive", "clustering_key_label": "S", "clustering_key_reasoning": "Monotonicity=0.854 indicates strong sequential ordering. This is the sequence axis within each drive, ordering plays chronologically. Ranges 1-20, organizing the timeline of each team's possession.", "world_line": "TeamWithPossession", "session_column": "DriveNumber", "confidence": "HIGH", "notes": "Primary perspective: partition by the team executing plays, ordered by play sequence. DriveNumber serves as session grouping within team timelines. This is an event table tracking individual play occurrences." }, { "rank": 1, "partition_key": "derived:concat(AwayTeam,'_vs_',HomeTeam,'_',Date)", "partition_key_label": "E", "partition_key_reasoning": "Alternative game-level perspective. Each unique game matchup becomes the entity. Would create ~3 partitions for this week's data (based on 3 unique dates). Useful for game-by-game analysis rather than team-centric views.", "clustering_key": "PlayNumberInDrive", "clustering_key_label": "S", "clustering_key_reasoning": "Same sequential ordering within each game context, maintaining chronological play order across all drives in a single game.", "world_line": null, "session_column": "DriveNumber", "confidence": "MEDIUM", "notes": "Alternative game-centric perspective. Useful for analyzing complete game narratives, but requires derived key. Less natural than team-based partitioning for typical NFL analytics." } ] } ```