Minimal Binary Multiple Windows on Odd Composite Survivors of Wheel 30

We introduce a binary-window framework for odd composite integers that survive wheel 30, that is, integers coprime to 2, 3, and 5. For such a composite integer n, we consider all contiguous binary windows in the binary expansion of n, reinterpret them as integers, and study the smallest window that is a multiple of a true factor of n. This defines a minimal multiple window Wmin(n) and induces a local/global dichotomy according to whether Wmin(n) < n or Wmin(n) = n. The wheel-30 restriction is used only as a domain reduction step. The central object is not a sieve in the Eratosthenes sense, but the arithmetic behavior of minimal binary windows on the reduced domain. Computationally, up to 10^6, every odd composite survivor of wheel 30 exhibited at least one factor-multiple window. Among the 188,170 composites in the tested range, exact visibility of at least one true factor occurred in 45.570495% of the cases. The dynamics split into a local regime, Wmin(n) < n, in 64.808418% of the cases, and a global regime, Wmin(n) = n, in 35.191582% of the cases. We also prove in Proposition 3 an exact prefix divisibility criterion for semiprimes. If n = pq with primes p < q, then for every k >= 1, p divides floor(n/2^k) if and only if p(q mod 2^k) < 2^k. This gives a complete characterization of prefix divisibility by the smallest prime factor. To study the smallest prime factor specifically, we introduce the refined observable Wmin^(p)(n), defined as the smallest binary window divisible by spf(n). In all tested cases up to 10^6, Wmin^(p)(n) existed for every composite survivor, and in all global semiprime cases the quotient Wmin^(p)(n)/spf(n) coincided exactly with the cofactor. In particular, among the 61,341 global semiprime cases found up to 10^6, this law held with empirical accuracy 100%.