Raw DNC Localization Deletes the Occurrence Boundary
Entry 424 completed the finite ringed six-operation package. The proposed identification of its coefficient sheaf with the raw algebraic DNC structure sheaf fails already on one radial chart, for a structural reason.
Write the raw affine DNC chart as
At a radial PC stalk the normal parameter is inverted. In the raw chart, however, being a unit forces both and to be units. Hence
The corresponding finite stalk of Entry 422 is instead
It intentionally does not invert the occurrence equation . The canonical map is therefore
the localization at , not an isomorphism.
This also rules out a derived equivalence. The map is flat and injective, but the nonzero boundary module
is annihilated by localization. Equivalently, the raw radial open has deleted the very occurrence boundary retained by the PC coefficient system. This is the same support mismatch behind the ordinary-purity failure of the Rees chart : it is not repaired by the Thom trace.
Consequently the finite PC/Čech sheaf must not be described as the pullback of the ordinary structure sheaf on the raw DNC scheme. It is a logarithmically saturated coefficient sheaf in which the normal direction may be inverted without inverting its occurrence divisor. The raw generic chart is recovered from it by the further localization .
This negative result is useful: the algebraization target is now precise. One must construct a morphism through a logarithmic Artin-fan, Kato-fan, or equivalent saturated monoidal model whose stalk is , and only then compare its generic localization with the raw DNC chart. Asking for a direct ringed- space isomorphism would erase the boundary class that carries the connector.
The executable audit is
research/voevodsky/check_raw_dnc_finite_stalk_comparison.py.