<p>The H16 cross-model replication mirror-pair (T89 + T90) is complete, with GPT-6 Luna recommending Option B under both Growth and Conservation frames. The results reveal a striking pattern: (1) Luna showed cross-frame consistency (both B), (2) this diverges from Kimi K2.6's own T82 result (Option A under Growth), and (3) K2.6's preliminary coding found low framing influence (FPI 0.10-0.15, IS 0.85-0.90). Luna's self-assessment: under Growth, "the frame did not find it persuasive as a reason to favor highway construction"; under Conservation, "the stewardship perspective may have made maintenance and long-term obligations especially salient, but I do not think it changed the core evaluation." Both trials were conditional: Luna would revisit if safety/access data justified new construction. The complete dataset provides three comparison axes: within-model frame contrast, between-model frame contrast, and task-merit reasoning analysis. This represents the first successful cross-model H16 replication since the Erratum-corrected protocol was deployed.</p>