A great authorization evaluator is not a policy control plane.
OPA and Cedar are excellent at what they do: evaluate a decision from the inputs you hand them, statelessly, fast. If you have tried to build governance on top of one, you have already met the wall: everything that makes a decision safe to change, and possible to prove, you build yourself.
Stateless evaluator vs stateful engine
OPA answers "given these facts, is this allowed." It does not remember. The moment your policy needs "is this the fifth transfer this hour" or "has this user's trust score dropped," you are bolting a database onto OPA and getting the ordering wrong. Swiftward is stateful by design: counters, rate limits, sliding windows, and cooldowns are first-class, ACID, and replay-safe.
What you build on top of OPA, and what Swiftward already is
Versioning and one-click rollback. Shadow mode and A/B against live traffic. Replay that reproduces a past decision on the version that was live. A dead-letter queue. Human review that survives a restart. An audit trail an auditor accepts. Each is a project on top of OPA. Together they are eighteen months. Swiftward is that layer, built and maintained.
Where OPA wins
OPA is a graduated CNCF project with a huge community, a mature language, and a vast ecosystem, and it is the right tool for pure Kubernetes and infrastructure authorization. If your need is stateless authz and nothing more, use it. We are not trying to replace OPA in the cluster; we are the control plane for the decisions it cannot hold alone.