Abduction as Premature Induction: A Formal and Game-Theoretic Defense

Abduction, properly construed in the Peircean sense as hypothesis nomination in response to surprising evidence, is not an autonomous inferential method but a strict subroutine of induction; treating abduction as justificatory without subsequent testing is This paper articulates and defends a containment view on which abduction launches inquiry by proposing candidates, while induction-via discriminating tests, replication, and convergence-alone licenses acceptance. We (i) situate abduction historically and conceptually with Peirce, Harman, and Lipton; (ii) formalize the containment claim with a Bayesian partition that makes explicit the penalizing roles of rivals and unconceived alternatives; (iii) model inquiry as a repeated game against Nature to show why myopic abductive "best responses" incur high regret absent exploration; and (iv) examine domains (e.g., miracle reports) where unknowns dominate priors and likelihoods, rendering abductive acceptance indefensible until testing collapses the catch-all mass. The result preserves abduction's indispensability as the first step of inquiry while disallowing its elevation to a rule of acceptance. has long been pulled between two projects. A broadly Peircean strand construes it modestly as the act of an explanatory hypothesis in response to recalcitrant observations (Douven, 2021; Peirce, 1931). By contrast, the Harman-Lipton tradition embeds abduction within a justificatory norminference to the best explanation (IBE)-on which explanatory virtues themselves underwrite acceptance (Harman, 1965; Lipton, 2004). This paper adopts the Peircean usage and argues that, under that usage, abduction is neither a third sui generis mode of reasoning alongside deduction and induction nor a direct warrant to believe. Rather, abduction is stage whose safety depends entirely on its feeding-forward into testing. Two families of arguments support the thesis. First, a Bayesian partition shows structurally why abductive acceptance is premature: without testing that reduces the posterior weight of rivals and the residual mass of unconceived alternatives, the numerator favored by an elegant hypothesis is outpaced by a stubborn denominator.