API Reference Matrix

更新日: 2026-07-05

主要 API の import path、引数、戻り値、副作用、関連テストを一覧化する。詳細な説明は各 module docs を参照する。

App CLI / Config

API	Arguments	Returns	Side Effects	Tests
pca.cli.main	argv | None	exit code	delegates to subprocess command	tests/test_app_config.py
pca.cli.delegated_command	argv	command argv	reads profile YAML	tests/test_app_config.py
pca.app_config.load_app_profile	path	AppProfile	reads YAML	tests/test_app_config.py
pca.app_config.args_mapping_to_argv	args mapping	list[str]	none	tests/test_app_config.py

Decision

API	Arguments	Returns	Side Effects	Tests
pca.decision.policy.deterministic_token_policy	obs	PolicyDecision	none	tests/test_encoder.py
pca.decision.policy.neural_policy	obs, model, torch, ...	PolicyDecision	optional runtime stats/cache update	tests/test_encoder.py
pca.decision.policy.neural_policy_batch	observations, model, torch, ...	list[PolicyDecision]	optional runtime stats/cache update	tests/test_encoder.py
pca.decision.remote_policy.remote_policy	url, timeout	Callable[[Any], PolicyDecision]	HTTP request per decision	integration/smoke

ISMCTS

API	Arguments	Returns	Side Effects	Tests
pca.search.ismcts.ismcts_policy	obs, your_full_deck, opponent_prior_deck, policy_fn, search_api, config, ...	PolicyDecision	CABT search sessions, runtime stats	tests/test_encoder.py
pca.search.ismcts.ismcts_policy_with_hidden	obs, hidden, policy_fn, search_api, config, ...	PolicyDecision	CABT search sessions, runtime stats	tests/test_encoder.py
pca.search.ismcts.candidate_action_indices	obs, node, encoded, limit, ...	list[int]	optional runtime stats	tests/test_encoder.py
pca.search.ismcts.visit_distribution	node, action_count, temperature	list[float]	none	tests/test_encoder.py
pca.search.ismcts.leaf_value_from_decision	obs, decision, root_player_index, value_config, stats	float	optional stats update	tests/test_encoder.py
pca.search.ismcts.finish_search_session	search_api	None	best-effort search_end()	tests/test_module_boundaries.py

Belief and Search Values

API	Arguments	Returns	Side Effects	Tests
pca.search.belief.sample_hidden_state	obs, your_deck, opponent_deck, rng, ...	HiddenStateSample	rng advances	tests/test_encoder.py
pca.search.belief.PublicKnowledgeTracker.update	obs	constraints object	tracker state update	tests/test_encoder.py
pca.search.mcts.search_value_config_for_profile	name	SearchValueConfig	none	tests/test_encoder.py
pca.search.mcts.terminal_value_for_player	result, player, result_reason, value_config	float | None	none	tests/test_encoder.py

Self-Play

API	Arguments	Returns	Side Effects	Tests
pca.training.selfplay.run_selfplay_battle	api, deck0, deck1, ...	(list[SelfPlayRecord], SelfPlaySummary)	CABT battle API calls	tests/test_encoder.py
pca.training.selfplay.collect_selfplay_games_parallel	CLI-derived args	(records, summaries)	worker processes, optional JSONL append	tests/test_module_boundaries.py
pca.training.selfplay.make_pending_record	obs, decision, ...	PendingRecord	none	tests/test_encoder.py
pca.training.selfplay.finalize_records	pending, result, ...	list[SelfPlayRecord]	none	tests/test_selfplay_summary.py
pca.training.selfplay.write_records_jsonl	records, path	None	writes JSONL	tests/test_encoder.py
pca.training.selfplay.append_records_jsonl	records, path	None	appends JSONL, writes lock file	tests/test_selfplay_summary.py

Training Data

API	Arguments	Returns	Side Effects	Tests
pca.training.dataset.load_records_jsonl	path	list[SelfPlayRecord]	reads file	tests/test_encoder.py, tests/test_training.py
pca.training.dataset.iter_records_jsonl	path	iterator	reads file lazily	tests/test_training.py
pca.training.dataset.collate_search_batch	records, ...	SearchBatch	tensor allocation	tests/test_training.py
pca.training.dataset.collate_belief_batch	records, ...	BeliefBatch	tensor allocation	tests/test_training.py
pca.training.dataset.record_policy_weight	record, ...	float	none	tests/test_training.py

Policy/Value Training

API	Arguments	Returns	Side Effects	Tests
pca.training.train.train_policy_value	input_path, output_path, config, best_output_path	metrics dict	writes checkpoints	tests/test_training.py
pca.training.train.policy_value_loss	output, batch, ...	(loss, metrics)	none	tests/test_training.py
pca.training.train.resolve_model_config	checkpoint/input/config args	dict	may inspect JSONL/card data	tests/test_training.py
pca.training.train.load_compatible_model_state	model, state_dict, ...	load report	mutates model weights	tests/test_training.py
pca.training.train.load_unified_feature_tables_from_jsonl	paths, ...	feature tables	optional cache write	tests/test_training.py

Belief Training

API	Arguments	Returns	Side Effects	Tests
pca.training.belief_train.train_belief	input_path, output_path, config, best_output_path	metrics dict	writes checkpoints	tests/test_module_boundaries.py, tests/test_training.py
pca.training.belief_train.belief_loss	output, batch, ...	(loss, metrics)	none	tests/test_training.py
pca.training.belief_train.belief_dataloader	paths, config	iterator	reads JSONL lazily	tests/test_module_boundaries.py

Evaluation

API	Arguments	Returns	Side Effects	Tests
pca.evaluation.tournament.run_head_to_head	deck/policy config	match summaries	CABT battle API calls	tests/test_encoder.py
pca.evaluation.tournament.run_against_deck_pool	deck dirs/config	evaluations	CABT battle API calls	tests/test_encoder.py
pca.evaluation.tournament.summarize_matches	match summaries	aggregate summary	none	tests/test_module_boundaries.py
pca.evaluation.tournament.write_csv_summary	summary/path	None	writes CSV	tests/test_module_boundaries.py

Rule Agents

API	Arguments	Returns	Side Effects	Tests
pca.rule_agents.registry.load_rule_agent_registry	path	RuleAgentRegistry	reads YAML	tests/test_rule_agents.py
pca.rule_agents.policy.rule_pool_policy_factory	registry/deck config	policy factory	may instantiate agents	tests/test_rule_agents.py
pca.rule_agents.generic.GenericAdvancedHeuristicAgent	observation methods	rule decision	none	tests/test_rule_agents.py

Compatibility Rule

公開 API を facade から出す場合は、tests/test_module_boundaries.py に以下を追加する。

• facade object と real module object の assertIs
• real object の __module__ 期待値
• 旧 import path を維持する必要がある場合の negative check