NULL: the methylation substrate does not mark hibernation either

RH8 closed one static sequence layer (stacking-stiffness γ), but γ is almost entirely GC-loaded here (cross-cell r(γ,GC) = 0.9955), so a skeptic could say only GC was tested. The methylation SUBSTRATE -- CpG observed/expected, which is GC-NORMALIZED -- is the natural second layer. Across the same 8 genes the number whose CpG O/E cleanly separates hibernators is 0. CpG O/E is a distinct read from γ (r = 0.5601) and far less GC-loaded (r(CpG O/E, GC) = 0.4927). A second, independent static layer is also blind to who can hibernate.

RH9, the v0.5.0 regulatory-layer object. The firewalled next step the v0.4.0 handover named: test whether the present-but-silenced gating is visible in a regulatory-relevant read, γ still reading structure only. CpG observed/expected is the genomic SUBSTRATE on which DNA methylation acts; it divides the observed CpG count by the count expected from C and G, removing the first-order GC dependence that drives the RH8 confound. It STILL does not separate hibernators -- so neither the stacking-stiffness layer nor the methylation-substrate layer encodes hibernation capacity. The capability is DYNAMIC regulatory gating of genes present in all (humans included), readable only in an in-vivo torpor↔euthermia methylation/expression contrast (cited, [O] external).

Why a second layer, and which one

RH8 (§11.1) showed no promoter γ marks hibernation, but every group γ gap there WAS the group GC gap. γ is a nearest-neighbour stacking-stiffness read and, across this panel, is almost entirely GC-loaded (cross-cell r(γ,GC) = 0.9955). The honest objection is that RH8 only ruled out GC. CpG observed/expected (Gardiner-Garden & Frommer 1987) answers it: it is the methylation SUBSTRATE -- the CpG-island architecture methylation acts on -- and it is GC-NORMALIZED by construction (observed CpG divided by the C,G-expected count), so it is NOT a restatement of GC.

The reads, by gene

For each gene: the hibernator CpG O/E range, the non-hibernator range, whether they OVERLAP (they all do), the hibernator-minus-non difference in mean CpG O/E, and an exact permutation p (exhaustive over all label splits, no RNG).

A distinct layer, and still blind

CpG O/E is genuinely a different read from γ: the cross-cell correlation r(γ, CpG O/E) is 0.5601 -- clearly sub-unit -- and CpG O/E is far less GC-loaded (r = 0.4927 vs 0.9955 for γ). Yet it STILL fails to separate hibernators (0 of 8 genes), and no gene survives a Bonferroni correction. Two independent static sequence layers -- stacking stiffness and methylation-substrate architecture -- are both flat across hibernation status.

Where the capability actually lives

The present-but-silenced gating is DYNAMIC: which CpG sites are methylated, when, and which transcripts move across the torpor↔euthermia cycle. That is a Layer-2 in-vivo measurement (cited: hibernation expression atlases and torpor methylation dynamics), NOT a static promoter property and NOT derivable from any sequence read. It is graded [O] external: the package's offline-reproducibility invariant precludes vendoring processed in-vivo data, so this is the named honest next step -- not a stub, and not faked here. γ still reads structure only.

Grades

The CpG O/E reads, the γ↔CpG-O/E dissociation, and the GC-loading contrast are [V]: reproducible offline and parameter-free. The conclusion that NO static promoter read marks hibernation is [O]: small n, phylogenetic non-independence, and the dynamic regulatory gating that DOES carry the capability is cited external biology, not derived here.