What Wins vs. What Sells: Knowledge Discovery in Disney Lorcana's Competitive and Collector Markets

Abstract

This study applies a Knowledge Discovery in Databases (KDD) pipeline to Disney Lorcana, a trading card game with a dual-market ecosystem driven by competitive strategy and Disney intellectual property collector demand. We integrate 2,652 cards from the Lorcast API (2,075 after deduplication), 749 tournament decklists from 710 unique players across 21 official Disney Lorcana Challenge and Community Championship Qualifier events in 13 countries, and three independent price sources spanning May 2025 through March 2026.

Our dual-axis SHAP framework reveals that rarity is the #1 predictor of market price but was not selected by Mutual Information filtering for competitive prediction — the competitive and collector markets operate on fundamentally different value systems. Deck build cost does not predict winning (r = −0.092, p = 0.012): Dogs (Amber/Emerald) costs $179 median and achieves a 24.8% Top-8 rate, while Blurple (Amethyst/Sapphire) costs $459 and wins only 18.8%. Chi-squared testing confirms archetype choice does not significantly predict winning (χ² = 11.85, p = 0.540) — card selection within archetypes matters more than which archetype a player chooses.

We introduce the Sleeper Card Index (SCI), a tournament-based metric bridging competitive performance (play rate × win rate) with market price, identifying buy/sell signals per archetype. The near-random predictive accuracy of our best model (XGBoost AUC = 0.544) is itself evidence of game balance: if deck composition alone determined outcomes, the game would be solved. The analytical framework — dual-axis SHAP with card-level SCI — is transferable to any TCG with competitive and secondary market data.