The Measure Problem for Observers: Probability, Typicality, and Self-Location Without a Global Sampler

This is the eleventh paper of a series that redescribes physical reality as a causally consistent history of information updates, and the second panel of its second triptych — observer, then probability and typicality, then epistemic trust — ordered by dependence. The tenth paper defined a record-bearing observer as an operationally individuated physical subsystem that acquires and integrates records, decodes them through an implemented internal channel, and lets the decoded content shape later processing. It defined that subject but did not say how to count or weigh it, and it handed forward five unresolved demands: plural individuation, the observer-grain problem, temporal segmentation, one-to-many lineage continuation, and the fact that a Boltzmann brain is not excluded from observerhood. This paper takes up the weighing. It does not propose the correct cosmological measure; it asks what must be specified before a sentence about typical observers or self-locating credence is even well posed. The contribution is a measure-specification problem, not a measure. Two theses organize the answer. First, the several things called “probability” — dynamical transition probability, the Born weight, empirical frequency, an observer selection measure, self-locating credence, and normative trust — are conceptually and type-theoretically distinct, and identifying any one numerical assignment with another requires an explicit theoretical or interpretive bridge rather than following automatically. Second, a third-personal typicality claim and a first-personal self-locating credence claim are specified separately, the former needing no self-location bridge. Formally, history is the base measure layer, weighted within a fixed decoherent-history framework; observer architectures, observation episodes, observer-moments, record-derived states, and lineages are within-history fiber types, weighted by a non-negative measure kernel concentrated on a type-specific eligible set. The base measure and the kernel are separately specified and independently motivated inputs whose factorization cannot be recovered from the final observer measure alone. Typicality is a reference-class conditional prevalence; self-locating credence requires a further bridge that reads a physical or selectional measure as first-person credence; a conditioning rule for a measure-zero conditioning event is distinguished from, and ordered before, the regularization prescription for an infinite measure; and causal provenance, veridicality, and channel reliability are kept apart, with provenance prevalence treated as a diagnostic, not a verdict. The account is positioned explicitly relative to the xerographic-distribution program of Hartle and Srednicki. The paper makes no independent empirical prediction; it derives neither quantum theory, the Born rule, nor a unique cosmological measure, and it offers no normative theory of trust. It supplies a typed specification, the minimal conditions under which a typicality statement carries meaning, and a hierarchical specification family over which the twelfth paper’s self-undermining-trust criterion can range.