Data-Driven Strategy: Scouting, EPA/OPR, and Alliance Selection

The best robot doesn't always win — the best alliance does. Elimination rounds are played by alliances chosen during alliance selection, so a team that scouts well and picks smart punches far above its own robot. This is the deep-dive that rookie teams most often skip and most regret skipping.

Two data sources, used together.

• Quantitative metrics you can pull instantly: OPR (Offensive Power Rating) uses linear algebra over alliance scores to estimate each team's average point contribution; The Blue Alliance publishes OPR per event. EPA (Expected Points Added), from Statbotics, models how much a team adds to an average match and is broken into component EPAs (auto, teleop, endgame). Statbotics' own analysis shows EPA performing well as a predictor relative to Elo and OPR, but it is still only a model.
• Your own scouting data, which captures what models can't: did their intake jam? Can they actually climb the Tower to Level 3, or only Level 1? Are they a reliable auto-scorer? Do they play defense well? This qualitative read is decisive in close picks.

Build a scouting system. Assign students to record, every match, the things that map to REBUILT scoring: Fuel scored in auto vs teleop, whether they LEAVE the starting zone in auto, endgame Tower level reached (Level 1/2/3 are worth 10/20/30 teleop points, and a Level 1 climb in auto is worth 15), and reliability/defense notes. A shared spreadsheet or a scouting app aggregates it. Cross-check your numbers against TBA and Statbotics — when scouting and EPA agree, you have high confidence; when they disagree, investigate why (a team may have improved mid-event, which a season-long model lags).

Turn data into a pick-list. Honestly assess your own robot's strengths and weaknesses, then rank candidates by who complements you. If you're a strong Fuel scorer but can't climb, prioritize a reliable high-level climber to chase the Traversal ranking point (earned at 50 Tower points in a match) and endgame value. If you score in auto and they don't, weight auto reliability. Rank for both first-pick (best all-around partners) and second-pick (specialists or solid defenders) scenarios, because you may pick later than you hope.

The case study: the canonical alliance-selection win is a mid-ranked team that scouted relentlessly, identified two complementary robots others undervalued, and assembled an alliance whose combined scoring cleared rank-point thresholds no single robot could. Strategy and data are a subsystem too — and unlike a flywheel, it costs only attention to build.