auto-decorrelation

Cross-family verification for AI code review recruits AI models from different architecture families to eliminate single-model blind spots when validating load-bearing project changes. It catches flaws inherent to a single model's architecture by using distinct verifiers.

You can verify high-stakes infrastructure updates by recruiting cross-family AI sidecars to perform adversarial checks on load-bearing changes. This decorrelated verification identifies issues in gate infrastructure or onboarding scaffolds that a single model would miss.

Yes, you can use local GPU models and external CLIs for adversarial code checking through consent-gated sidecar recruitment. The system discovers available verifiers like Codex or Gemini to maximize family diversity against the orchestrator model.

When no cross-family sidecars are available, the system gracefully degrades to same-family in-session verification and provides honest notes about the limitation. It never hard-fails due to missing external tools.

Use multi-model verification when modifying gate infrastructure, onboarding scaffolds, or skills that produce trusted input for downstream components. These load-bearing changes require decorrelated checks to prevent architecture-specific flaws from propagating.

No, the governor requires source-grounded acceptance of all sidecar findings during cross-family checks. This ensures that every verification result from external CLI recruitment is validated against actual project sources before integration.