the differentiator
A policy is trusted only once it survives traffic it never saw.
Every incumbent learns least privilege from observed traffic. It scores perfectly on the window it watched, and silently blocks the failover. This is the part of Themis that exists because that is not good enough.
0100 the held-out test
Learning from observed traffic is a trap with a name.
Learning a policy from observed traffic looks perfect on the traffic it watched. Then the failover runs, the nightly batch runs, someone breaks glass at 3am — and the flow that was never in the window gets dropped.
So the policy is scored against a held-back set and an adversarial set before it goes anywhere near enforcement. A pass enforces on one slice and widens only by declaration. A fail returns a gap list, never an outage.
The separation is mechanical, not procedural: the drafting stage has no path to the held-back set, and that is checkable by walking the closed set of files it opened. It is not a promise that someone remembered to be careful. That is what makes a pass mean anything.
Observed traffic is evidence, never authority.
0200 limits
What this does not do.
e000 End.DT6 deliver
See it against your own kit.
30–60 minutes. Read-only. We compare your declared intent to what is actually running, and hand back the gap list.
Nothing is installed to have this conversation.