MultiEngineSynthesis

Formal methodology for consolidating independent assessments, reviews, or research from multiple PAI engines (PNC, PNK, PNX, Gemini) into a single authoritative output.

Thesis

Four independent AI engines assessing the same target produce complementary blind spots, different frameworks, and orthogonal coverage. The methodology captures a repeatable pattern for reconciling these: DISCOVER → LOAD → COMPARE → RESOLVE → CONSOLIDATE → VERIFY → APPLY. Each engine’s unique findings are preserved — consolidation is not flattening.

Evidence

Proven on the Altana Product Passport assessment: 4 engines, 10 resolved contentions (K1-K10), 513-line consolidated output with 20 risks, 9 sections. Each engine contributed unique findings that would have been missed in a single-engine assessment.

Key Design Decisions

  • Canonical spine selection: The engine with the broadest framework coverage becomes the backbone. STRIDE (PNC) subsumed custom categories (PNK).
  • Precision overlay: Qualitative scoring primary (more honest), numeric overlaid for cross-referencing.
  • Actor-actionable routing: Vendor-side capabilities become DDQ items, not customer mitigations.
  • 6 resolution rules in priority order: superset > precision > specialization > actor-actionable > conservative > escalate.

Implications

  • Multi-engine consolidation is not optional for thorough assessments — no single engine covers all dimensions
  • Skills directory at ~/.opencode/skills/MultiEngineSynthesis/ (PNK tree, 8 files, ~1133 lines)
  • Algorithm integration documented at AlgorithmPattern.md as mode: synthesis sub-phase (4b/7)
  • Applicable to security assessments, code reviews, and research synthesis