Frozen after the main submission, before pooling controls.

Question: can the new uniform one-sided lower count turn known calibration
of a slightly larger parent interval into unweighted conditional transport
on almost all smaller disjoint windows?

N-child windows, M children, common n and model massp_n. q_j their actual
selection masses, meanq=q_parent exactly. Uniform local calibration theorem
gives q_j/p>=1-delta_N, delta_N=selectedprecisiontail+2^k/N.
If eta=|q_parent/p-1|, conjectured deterministic bound:
 average_j TV_j <=eta+2delta_N.
Thus most windows calibrate when both errors vanish. For M=ceil(N^(1/50)),
known parent decoder regime includes all local c<c_cap, since
c_cap/(51/50)<c_dec. This is local density-one transport, NOT uniform
transport for every individual child and NOT a larger horizon measured
against the full parent source-bit length.

Check all finite vectors of ratios in fixed rational menus, including
the negative control(1/2,3/2) showing pooled calibration alone is insufficient
without the uniform local lower bound. Verify the threshold inequality
using the certified rational enclosures. <=60seconds CPU, no actual
enumeration at symbolic N^1.02 and no MonteCarlo.
