Are we typical observers? Cosmology's typicality arguments — Boltzmann-brain domination among them — answer by counting: observers of our kind against observers overall, under some cosmological measure. The count has two silent presuppositions. First, that some canonical, universe-wide observer count is available at all — this paper's foil, the Global Sampler; second, that all counted observers stand on equal evidential footing, their memories weighted alike whether or not those memories were ever acquired. The Boltzmann-brain problem (Dyson, Kleban its cutoff-measure form, De Simone et al. 2010) is acute precisely because a thermally fluctuated brain is an observer — this series' Paper 10 registered the rule that excluding it by fiat "would circularly dissolve the problem" — and yet its "memories" were never acquired from anything. Paper 11 retired the abstract sampler (a mathematical measure entails no physical sampler); Paper 37 barred its cosmological reification (the measure a typed specification, adopted not canonical; Boltzmann-brain domination diagnosed as "self-undermining = cognitive instability, not falsity"). What survived both is the counting practice's second presupposition: warrant-flat evidence. This paper — the closing panel of the Reconnection Program (Papers 60-62) — types it. The construction is a warranted-evidence fiber: Paper 11's typed eligible set of observers, realized patch-locally on Paper 61's certified record ledgers, with each member's record profile graded on two orthogonal axes — acquisition provenance (does a represented content have a corresponding acquisition lineage? Paper 21's anti-luck grades, evaluable for private records) and export standing (Paper 29's provenance-admissibility where records were exported, with "no export" a neutral value carrying no negative weight). A grade-bridge lemma fixes the codomains and the profile-aggregation rule; a sampler-deformation test — over bounded, strictly positive, support-preserving reweightings of Paper 11's base-measure and kernel factors, which change normalized stratum prevalences (a concrete mover is exhibited) while fixing every fiber-local datum — certifies that membership, grades, and strata consume no count; and the test has a genuine casualty: a prevalence-conditioned grade (standing conditioned on the cosmic frequency of one's stratum) moves under the class and is typed to the claim layer, where Paper 11 always kept such content. The Boltzmann-brain stratification theorem then says exactly this, with Paper 10's non-exclusion rule quoted in its statement: the fluctuated observer stays in the reference class; its instantiation profile fails the acquisition axis (pseudo-memories: represented provenance with no acquisition lineage); an ordinary profile does not; so the comparison is never warrant-flat in any admissible fiber, and counting it warrant-flat is a typed, priced, admissible choice (a provenance-blind kernel), never a forbidden one. A persisting fluctuated observer that acquires records ordinarily has them graded normally — the stratification is profile-relative, not essentialist. The measure-spec classification types the four declared reference specifications (proper-time cutoff; scale-factor cutoff; causal patch; causal diamond — as enumerated in the cited sources; not an exhaustive taxonomy) on two decision-ruled axes: none, as specified there, factors its count through warrant strata — the practice's measures are uniformly provenance-blind, a finding priced by the theorem's corollary, with an add-only row open for a provenance-respecting successor; no specification is selected, ranked, or scored, and the cosmological measure problem is not touched. A strict-recovery lemma returns Papers 11 and 37's registered wordings unchanged. ReconStatus(typicality) = scoped at model grade; open by seat: measure selection, Boltzmann-brain frequencies, anthropic or doomsday-style inference, credence bridges beyond Paper 11's declared one, and every physics row. The grouping's closure ledger reports all three Reconnection panels and names the residue once, each row owned. No accounting-layer adoption; no calibration; no datum; the substantive sampler face is declined, not refuted.
Tomoyuki Uchida (Mon,) studied this question.