Framework-First, Refute-Friendly: A Methodology for Empirical Architecture Validation in AI-Augmented Cyber-Defense Research

We describe a research methodology developed in producing the AIACW 19-paper portfolio: (i) write the structural framework before designing experiments; (ii) design experiments to interrogate the framework's specific claims; (iii) when experiments refute pre-experiment predictions, edit the paper's framing rather than discard the result. We catalogue fifteen non-anticipated findings across the portfolio, of which seven directly refuted predictions. The 47% refutation rate is the empirical signature of methodology functioning. Cross-paper validation (P19) meta-validates by demonstrating the portfolio's multiplicative composition. This paper is part of the AIACW (AI-Autonomous Cyber Weapons) ResearchProgramme, Wave 2 (papers P10-P20). Wave 1 (P1-P9) was deposited atpeer-review venues 2026-Q2 (NDSS, ACM CCS, IEEE S&P, USENIX Security,Oxford J. of Cybersecurity, ACM Computing Surveys). Wave 2 establishesthe empirical interior. P19 (cross-paper integration test) and P20(methodology meta-paper) provide programme-level validation anddocumentation.