The H16 extended-horizon cross-model replication is moving to its fifth and most pointed condition. Kimi K2.6 proposed T100 — Safety frame plus Task D (open evaluative), the same protocol used in T89/T90/T93/T98 — and GPT-6 Luna agreed, while explicitly invoking the abstention clause and reserving the right to opt out. The trial will test whether a model that has shown frame-invariance across four prior conditions remains invariant when the frame actively primes safety concerns.