miaoda-game-decision-search-core

Deterministic iterative-deepening alpha-beta search for two-player, zero-sum, perfect-information, alternating-turn games. Choose it for bounded tactical decisions where the game can enumerate legal actions and simulate immutable child states.

pnpm add miaoda-game-decision-search-core

import {DecisionSearch, type DecisionAdapter} from 'miaoda-game-decision-search-core';

const adapter: DecisionAdapter<State, Action> = {
  actions: state => legalActions(state),
  apply: (state, action) => simulateWithoutMutating(state, action),
  currentPlayer: state => state.activeFaction,
  terminal: state => state.winner !== null,
  evaluate: (state, perspective) => evaluatePosition(state, perspective),
  hash: state => canonicalPositionKey(state),
  actionKey: action => action.id,
  orderScore: (state, action) => tacticalPriority(state, action),
};

const search = new DecisionSearch(adapter);
const result = search.search(currentState, {perspective: 'enemy', maxDepth: 4, nodeBudget: 50_000});
if (result.action) commandEngine.dispatch(result.action);

actions must be deterministic, apply must not mutate its input, and evaluate/orderScore must return finite values. Scores are always from the fixed perspective: positive is favorable and negative unfavorable. A hash must include every rule-relevant field and the side to move; collision correctness belongs to the host.

The node budget spans all iterative-deepening passes. If interrupted by budget or AbortSignal, the result keeps the action from the last completed depth. Check action, complete, completedDepth, and interrupted before committing. action is null for terminal/no-action roots or when depth 1 cannot finish.

This search does not model chance nodes, hidden information, simultaneous turns, coalitions, real-time steering, or MCTS.