Is quiescence search related?

Yes. Seeking a more stable position before static evaluation reflects an important part of Shannon's proposal, although individual implementations vary.

Is type B guaranteed to beat type A?

No. Its benefit depends on selection quality, evaluation, available resources and the positions encountered.

Does a modern engine have to belong purely to one type?

No. Contemporary designs can combine broad search, selective extensions, reductions and other mechanisms; the historical labels need not describe every detail.

AI Glossary / Graphs and symbolic AI / Intermediate

Type B strategy

Definition

A type B strategy is Claude Shannon's historical approach to game search that selects which variations to investigate and extends important lines toward suitable evaluation points. It uses judgment about relevance and position stability to allocate search effort rather than treating all lines uniformly.

What it means

Shannon introduced the distinction in his 1950 paper on computer chess. Type A considers all variations to a definite depth. Type B combines selective exploration with attention to whether a position is sufficiently settled for an approximate evaluation.

The appeal is practical: some branches are strategically important, while others may not justify equal computation. A forcing sequence can warrant deeper investigation because each move creates a narrow set of urgent replies. A quiet-looking move can nevertheless matter, making selection a source of risk.

Type B is a historical design idea rather than one complete modern algorithm. A system claiming to use it still needs explicit rules for selecting moves, extending lines, stopping search and evaluating positions.

How it works

A move-selection procedure identifies variations considered worth exploring. A second judgment determines whether the current position is suitable for evaluation or whether unresolved tactical activity calls for further search. These judgments can be implemented through domain-specific functions and resource limits.

Selected branches are examined and their evaluations combined according to the game-search model. Because omitted branches may contain a decisive reply or a strong move, selective search generally loses the simple coverage guarantee of a full-width search at the same horizon.

This should be distinguished from sound alpha-beta pruning, which omits work only when bounds establish that it cannot change the corresponding minimax decision. A heuristic decision that a move looks unimportant does not provide that proof. Extensions also need limits: attempting to resolve every active sequence can create excessive or poorly controlled computation.

Three practical examples

A chess-programming exercise extends capture sequences before applying a static evaluation, illustrating the need for a suitable stopping point.
A board-game agent ranks candidate moves and explores only a subset, then tests whether the selection misses important defensive replies.
A search researcher compares selective policies with a full-width baseline under the same total computation budget.

Think of it this way

Think of an analyst spending more time on a few important scenario branches and continuing an unfolding sequence until its immediate consequences are clearer. This can reveal detail that uniform shallow review would miss.

The analogy does not make the selection infallible. A branch dismissed early may contain the crucial outcome. In game search, the selection and stopping rules must be evaluated as algorithms, not accepted because they resemble expert attention.

Worked example

The cost of omitting one reply

Consider an invented root choice between moves A and B. After A, the opponent has replies with boundary values 8 and −6 from the root player's perspective. After B, the replies have values 3 and 2.

A full two-ply minimax calculation gives A the value −6 and B the value 2, so it chooses B. Suppose a selective procedure excludes A's second reply because it initially looks unpromising for the opponent. It then assigns A the value 8 and chooses it, missing the represented refutation.

Now consider the separate stopping issue. A boundary position is initially scored +4 after a capture, but a forced recapture one ply later changes the estimate to −2. Extending that active line can correct the premature evaluation.

The two examples illustrate different parts of the type B idea: selecting useful lines and choosing appropriate evaluation points. Improving the second does not automatically repair an omitted branch in the first.

Illustrative case study

Fictional Willow Games develops a selective search policy that prioritizes checks and captures. It performs well on demonstrations built from forcing tactical sequences, but later misses a quiet defensive move.

The team adds tests where important moves lack obvious forcing features and compares the policy with a bounded full-width reference. It also records where extensions stop, so a high reported depth on one line is not mistaken for equally deep coverage of every alternative.

The fictional case shows why average speed and maximum depth are insufficient to judge selective search. It supplies no measured playing-strength result and does not claim that one selection rule is best across games.

Common misconceptions and mistakes

Type B simply means deeper search. It also concerns which variations are selected and where evaluation is appropriate.
Any pruning method has the same risk. Proven alpha-beta cutoffs and heuristic exclusions have different justifications.
An expert-looking move ranking cannot miss strong moves. Selection can discard subtle alternatives, sacrifices or necessary defenses.

Business use

The type B idea is useful when a product must make decisions under a strict time budget. It exposes the tradeoff between covering alternatives and examining selected alternatives in greater detail.

When evaluating a selective search system, ask which branches can be excluded and how those decisions are tested. Report the distribution of search depths, the resource limits and failures caused by omitted moves. In business simulations, ensure that the game assumptions fit the problem before importing chess-inspired search ideas. A deeper-looking explanation is not evidence that the important alternatives were considered.

Check your understanding

A selective search reports that it explored one line to depth twelve. Does this establish that every legal root move was analyzed to depth twelve?

Answer: No. Selective search can investigate different branches to different depths and may omit some branches entirely. The coverage must be reported separately.

Related terms

Learn next

Read quiescence search and alpha-beta pruning. Compare a branch omitted by a proven bound with a branch omitted by a heuristic, and explain what evidence supports each decision.

Sources

Claude E. Shannon, Programming a Computer for Playing Chess, 1950, Section 6 defining type B through selective exploration and appropriate stopping positions: https://www.computerhistory.org/chess/doc-431614f453dde/

Russell and Norvig, Adversarial Search, discussion of cutoff evaluation and forward pruning: https://aima.cs.berkeley.edu/4th-ed/pdfs/newchap06.pdf

The museum-linked original paper and relevant textbook passages were read. The numeric trees and fictional case are original illustrations. Historical terminology is explained without claiming that it completely classifies modern engines.

Previous: Type A strategy Next: Type-Casting