AI Policy Simulation

Changing a governance policy is risky if you cannot see what it will do first. EVE Shadow lets authenticated customers evaluate a candidate policy without changing the authoritative decision — live alongside production traffic or by historical replay — and see exactly where the candidate would diverge.

What shadow evaluation does

• Non-authoritative by construction — the candidate policy is evaluated in parallel; it cannot alter the active verdict.
• Live or replay — run the candidate against live traffic in shadow, or replay historical decisions against it.
• Signed divergence reports — get a signed report (jcs-1) showing where the candidate agreed and diverged, with modeled-effect estimates and divergence broken down by class.
• Dry-run promotion — promotion is dry-run only and never activates a production policy.

Evidence example

signed shadow report v2 (jcs-1) {
  agreement_permille,
  divergence_by_class,
  critical_regressions
}

Links

• Policy enforcement: /seo/landing/ai-policy-enforcement.md
• Adversarial regression testing: /seo/landing/ai-red-team-testing.md
• Determinism: /seo/landing/deterministic-ai-governance.md

Readiness

Shadow policy evaluation and replay divergence reporting are PILOT_READY; the shadow subsystem is authenticated pilot-ready (Python; hosted and embedded-service modes), covered by the shadow phase report (27 shadow tests).

Limitations

• Shadow evaluation is structurally non-authoritative and cannot alter the active verdict. Promotion is dry-run only and never activates a production policy.
• Divergence estimates are modeled effects from replay or sampling — they are explicitly not observed production outcomes.
• Shadow evaluation is available to authenticated customers.

Next step

Load a candidate policy into shadow during a pilot, replay your history against it, and read the signed divergence report before you promote anything.