How engines work

A representation with 64 zero-or-one indicators, one per square, marking which squares belong to a chosen set.

A procedure that assigns a numerical estimate to a position so a chess program can guide its search.

A neural-network architecture designed to evaluate chess positions while updating efficiently between nearby positions.

An algorithm that chooses a move while assuming both sides will always make their best available decisions.

A method that skips branches unable to change the final choice when both sides seek their best result.

A strategy that repeats search with increasing limits so a completed result is always available.

A technique that chooses which moves to search first so unhelpful branches can be rejected sooner.

An optimization that fully searches the first candidate and initially tests later moves with a narrower check.

A technique that initially searches moves placed late in a position's analysis order at reduced depth.

A heuristic that tests a fictional pass to discard positions that appear clearly sufficient.

An extension that continues selected tactical moves until the position is stable enough to evaluate.

An error caused when a decisive consequence lies just beyond the search limit and distorts the evaluation.

A method that repeatedly explores variations and spends more analysis on moves that appear promising.

A fast calculation that estimates the material balance of successive captures and recaptures on one square.

Memory that lets a chess engine reuse results for positions it has already searched.

A method that creates a quickly updatable numerical fingerprint for the complete state of a chess position.

A test that counts every legal move sequence to an exact search depth.