Frozen before literature search and calculations, 2November2026.

Hypothesis: ordered selected canonical types have typical cost Q_n/n->mu_B
under the centered tilt, already present in Collatz35/v2 proof. Each selected
word has Haar atom2^-Q/p_n. An actual selected ensemble has at mostNq_N
distinct words. If q_N/p_n is bounded/subexponential and n>log2N/mu_B,
its support cannot carry significant Haar selected mass. Hence conditional
TV should tend to1, not0. This would bound THIS joint relative-selection
transport target, not individual Collatz or all conditioning methods.

Need prove cardinality argument, establish selected typical cost from the
accepted tilt/normalizer, retain q_N/p_n qualifier, and check prior art.
Without relative selection control, generic M<=N only gives the weaker
ceiling1/(mu_B+log2lambda), not1/mu_B. Do not erase this distinction.

Finite controls: exact support-mass inequality against independent top-M
atom sums in a bounded actual type box; no finite tests imply the asymptotic
theorem. At most4 blocks with r,s<=4, then exact word probabilities, fixed
menus of M and costcutoff. One CPU process,60second cap.
