This is the twelfth paper of a series that redescribes physical reality as a causally consistent history of information updates, and the closing panel of its second triptych — observer, then probability and typicality, then epistemic trust — ordered by dependence. The tenth paper defined a record-bearing observer and established that such an observer's grip on its own records secures their downstream usability but never, with certainty, their upstream provenance (Uchida, 2026j). The eleventh paper specified what must be declared before a claim about typical observers or self-locating credence is well posed, and handed forward a hierarchical specification family together with a diagnostic, the provenance prevalence, explicitly marked as an input and not a verdict, with the cosmological measure, reference class, regulator, and self-location bridge held to be independent inputs rather than consequences of the physics (Uchida, 2026k). This paper supplies the missing normative step: the last arrow of the eleventh paper's chain, from self-locating credence to rational trust in records. Its central object is a reflexive consistency condition — not on a cosmology in isolation, but on a cosmology together with the observer, measure and self-location, process-adequacy, inferential, and normative specifications under which it is evaluated, that is, on a theory–specification package. Four commitments shape the construction. Trust is treated as warrant, the entitlement to use a record-derived state as evidence for its referent, and is analyzed in layers: first-order process adequacy (the evidence was in fact formed through a reliable, causally continuous network free of declared objective exclusions, a layer non-factive about token truth), a reflective adequacy credence (a first-person credence that process adequacy holds), and a reflective entitlement (the warrant proper, the verdict a warrant bridge confers on that credence and on any defeater the observer knows of). The content of evidence is separated from the physical token bundle and formation network a given centre actually carries, since centres that share an apparent evidential content can differ in provenance, reliability, and preservation. Evidence is in general a bundle, and its reliability is an aggregate over a whole evidence network, sensitive to redundancy and common-mode dependence. And the criterion, the Non-Self-Undermining Condition (NSU), demands consistency between theory-support and evidence-entitlement: in any applicable case — one in which the evidential content, under a declared inferential specification, supports the theory — the package must license the use of that evidence. The condition is given in majority, threshold, and admissible-family-robust grades, and is positioned relative to the xerographic-distribution program of Hartle and Srednicki and the cognitive-instability arguments of Albrecht–Sorbo and Carroll. The paper's scope is deliberately negative and bounded. NSU is necessary, not sufficient: in an applicable case, passing establishes package-relative reflective entitlement under the declared bridge, not process adequacy, and a fortiori not truth, rational acceptance over rivals, or the correct measure (No-Self-Certification); where the content does not support the theory the condition is simply not applicable, not passed. NSU does not refute external-world skepticism, since it governs only the reflective layer and lifts no observer out of the predicament that upstream provenance is never directly verifiable from the inside; a package that, by its own combined specifications, withholds entitlement from its own supporting evidence is not thereby false but cognitively unstable — cognitive instability, not refutation. It does not solve the bootstrapping problem, and it derives neither the credence threshold, the bridges, the correct cosmological measure, nor the Born rule. Its contribution is a typed, layered account of epistemic trust, a formal criterion that filters self-undermining theory–specification packages, and the closure of the second triptych from inside the history, with no view from nowhere to certify records.
Tomoyuki Uchida (Sat,) studied this question.