Module beam_search 

Parametrizable beam search for Quantik.

Descends level-by-level from a root state, keeping only the top beam_width non-terminal candidates per depth (breadth pruning) while always discovering and recording every true terminal state encountered, regardless of the beam width. Port of the Python quantik_core.beam_search module (schedules, multiplicity accounting, ranked root moves, wall-clock budget) — without the shared CompactGameTree: this engine is standalone, and nodes_inserted counts the terminal leaves recorded plus the survivors kept per level, the closest observable analogue of the Python tree-insertion counter.

Structs

BeamLeaf

A single collected leaf: a principal variation and its value.

BeamSearchConfig

Configuration for the beam search algorithm.

BeamSearchEngine

Level-by-level beam search over the Quantik game tree.

BeamSearchResult

Result of a beam search run.

BeamStats

Search statistics.

RankedRootMove

Aggregated beam-sampled statistics for one first move from the root.

Constants

UNIQUE_CANONICAL_STATES_PER_DEPTH

Unique canonical states per depth (see GAME_TREE_ANALYSIS.md in the Python repository). Useful for building an exhaustive-prefix beam_schedule that keeps every legal line up to some depth before switching to guided sampling.

Type Aliases

Evaluator

Evaluates a position from player 0’s perspective; values are clamped to [-1, 1] by the engine.