✶ Induction - The Best Game in Town - Performance, Rule-Circularity, and the Pragmatic Vindication of Expectation

Hume's problem is often presented as a forced choice: either induction can be proved without using induction, or reliance on it is merely faith. That choice is false. Deduction cannot establish that unobserved cases must resemble observed ones, but finite agents still need methods for forecasting, diagnosis, explanation, and action. The relevant question is comparative: which available method remains most accountable to what happens? This paper defends a conditional pragmatic vindication. Its bridge principle is modest: when methods face the same recurring task and information, and one corrigible method persistently incurs less purpose-relevant error without a countervailing cost, preferring the worse method requires a special reason. Proper scoring rules make confidence assessable; performance-weighted meta-induction reduces dependence on a lucky temporary leader; and no-regret results specify how a learner can remain close to the best accessible rival over time. Examples involving alarms, weather, navigation, medicine, home repair, manufacturing, spam, finance, and historical inference show how the comparison works and where it can fail. The result is narrower than a proof of universal uniformity. Performance depends on sample size, error costs, the score, rival set, horizon, and structure of the domain. Sometimes the rational output is wider uncertainty or abstention rather than a confident forecast. Even so, the absence of deductive proof does not place outcome-sensitive inquiry on an epistemic level with commitments insulated from every possible failure. It supports calibrated reliance: use a method only to the degree, in the domains, and for as long as its comparative performance warrants.