Second research track, frozen before the screen controls. Collision: central unfinished screen + relay exchange + end-state conditional freshness. Work at a deterministic bit horizon H with exactly k completed 01 transitions and optional last bit. Derive the exact Markov likelihood including the endpoint factor, bound the rarity after matching H/k, then reuse the proved early unweighted arithmetic interface for this different conditional law. For k=n-r at a central H, compute how a small missing count changes the survivor exponent, with no implication about subsequent actual survival.

Controls to run only after the identities are written: exact forward DP through H, classify every survivor by count and terminal bit, compare to an independent sum over complete positive run pairs and one partial block, using all H<=22 and both last bits. Compare literal Markov word mass to the transition-count likelihood in rational proxies. A bounded screen histogram may illustrate the distinct missing-count scales. No assessment of collision prompt efficacy.
