Skip to main content

Component: PDP

What it is: the Policy Decision Point — a pure function (policy + request + verified token in, decision out; no I/O of its own).

What it does: answers "can this caller call this target, with this data class, at this hop, given this taint?" against the session's pinned policy version.

Why it's pure: the PDP takes a policy, a request, and a verified token, and returns a decision — no database call, no clock read, nothing that could make the same inputs produce different outputs on a retry. Every side effect (recording the decision, minting a token) lives in the caller, which is what lets CI Drift Action run the exact same edge-classification logic offline, with no network call, and get the identical answer the runtime would have given.

What decides recoverability: recoverability is a property of the reason, not a per-call judgment — see Reference → Authorize schema for the full classification, so the PDP, the enforcement point, and the SDK can never disagree about which denials are worth a step-up prompt.