Fixedguess

Fixedguess is a term used in theoretical discussions of prediction and decision strategies to describe a predictor that outputs a single, fixed guess regardless of available evidence or input data. It is a degenerate or constant strategy that provides a simple reference point for evaluating more complex models.

As a constant function, fixedguess provides a baseline against which adaptive or probabilistic models are measured. In binary settings, the performance of a fixedguess option equals the prior probability of the chosen class; for example, always predicting class A yields accuracy equal to the base rate of A. This property makes fixedguess useful for illustrating the potential value of information and the limits of non-adaptive decision rules.

Fixedguess is not a practical predictor for real tasks, but it serves as an interpretable reference. It helps demonstrate how much predictive power is gained from incorporating evidence, learning, or uncertainty handling. In disciplines such as decision theory, game theory, and machine learning, fixedguess is discussed as a degenerate strategy that clarifies the role of information and strategy selection.

Common examples include weather forecasting scenarios where one might always predict rain, or spam filtering where all messages are labeled as not spam. The concept emphasizes that without supporting evidence, any fixed decision has limited success and is easily outperformed by strategies that use data or feedback.

See also: baseline predictor; constant predictor; zero-information strategies.