Shadow mode for policy: simulate before you enforce
May 8, 2026 · 6 min read
Shipping a new control to production blind is how you take down a customer workflow. Replay real and synthetic traffic against a candidate policy first.
Nobody deploys a database migration without running it against a copy of production first. Policy deserves the same discipline. A control that's slightly too aggressive doesn't just annoy users - it silently blocks a legitimate workflow, and you find out from a support ticket.
Simulation runs a candidate Edict against real or synthetic traffic and reports exactly what it would allow, redact, block, or escalate - without affecting a single live request. You see the blast radius before you ship it.
Synthetic traffic is where this gets interesting. We generate adversarial inputs - PII in unusual formats, tool calls phrased to slip past a naïve rule - and replay them through the candidate policy. It's red-teaming your governance before an attacker does.
Shadow mode turns policy changes from a leap of faith into a reviewable diff with a measured impact. That's the difference between governance you trust and governance you fear touching.
Further reading
