✶ Base Rates, Identity, and Statistical Resistance: Why Prior Probability Feels Like Prejudice When It Threatens the Self

The Credencing framework evaluates how assigned confidence answers to evidence. Its existing papers distinguish objective evidence, perceived evidence, assigned credence, and deep rationality; separate evidence-processing failures from belief-integrity failures; defend warranted uncertainty; diagnose rationalization; extend credencing to institutions and AI-mediated inquiry; distinguish raw from excess gaps; make posteriors answerable to evidential history; credential frames and categories; and distinguish compatibility from confirmation through likelihood literacy. This paper adds the prior-probability competence that remains underdeveloped in that web: base-rate literacy. A base rate is disciplined statistical memory: structured information about how often a kind of event, condition, outcome, or behavior occurs within a specified reference class. Agents often reject such information not because it is irrelevant, but because it feels morally flattening, identity-threatening, hope-destroying, fear-amplifying, or socially dangerous. The lacuna is therefore not ordinary base-rate neglect alone. It is statistical resistance: the motivated refusal to let relevant background frequency constrain prior confidence when doing so would threaten self-conception, group identity, preferred narratives, or emotionally needed exceptions. This paper formalizes reference classes, base-rate prior pressure, reference-class confidence, prior constraint, individuating evidence, warranted base-rate slack, base-rate gaps, and excess statistical resistance. It argues that rational base-rate uptake requires neither blind deference to aggregates nor personal exemption from aggregates. It requires accountable movement among three questions: which reference class is apt, how strong is the base-rate signal, and what individuating evidence licenses departure from it? The practical result is a base-rate audit protocol for individuals, institutions, and AI-mediated inquiry: state the claim, identify live reference classes, estimate their base rates, assess category confidence, separate base rates from stereotypes, register individuating evidence, preserve uncertainty where classes are unstable, and diagnose residual resistance only after warranted slack has been granted. Credencing thereby gains a missing prior-discipline: confidence should not begin wherever identity, fear, or hope would like it to begin.