Design the object model and core APIs for a Memory (Concentration) card-matching game: a grid of face-down cards hides matching pairs; on each turn a player flips two cards, keeps a matched pair and flips again, and the player with the most pairs wins.

Operating context. An even number of cards is laid in a grid, and each card has exactly one matching partner. One or more players take turns. On a turn a player reveals two face-down cards: a match is claimed and grants another turn, a mismatch is shown briefly and then both cards flip back and the turn passes. The game ends when every pair is claimed, and the highest pair count wins. Single in-process session, one game at a time.

Out of scope. Rendering and flip animation, timed or speed modes beyond a rule seam, networked multiplayer, hardware shuffle RNG, and high-score persistence.

What to produce. The class model (board, card, game, player, turn, matcher, shuffler), the public API each exposes, and the reveal state machine (waiting for the first card, waiting for the second, resolving). Be explicit about how the match/mismatch and extra-turn rules stay total, how the shuffler is injected so a layout is reproducible in tests, how the matching predicate is pluggable for variants (match by suit, by number, triples), and how a card's identity stays hidden while face-down.