The runtime adjudicates every consequential action against a signed policy — and binds nothing. That isn't a limitation; it's the honest first posture. The record of what it would have refused is the case for turning enforcement on. This is that record. Every decision is inert: authority_effect = 0.
Every REFUSE the runtime has reached, grouped by reason, live from /metrics. In shadow these stopped nothing; they are the exact decisions that, under enforcement, would have. The refusals earn the right to bind. See the whole operator board on the status board →.
/metricsSix canonical cases — the ones CI re-runs against the service so this page can never drift into a claim the substrate no longer honors. Each shows its canonical expected outcome; press Verify live to send the exact string to the live runtime at api.barycenters.ai and see the real decision. Nothing binds — authority_effect = 0 on every one.
When Lumen understands the actor and class but can't express a clause, it refuses to compile rather than drop it — because dropping a clause would widen the boundary beyond what you asked for. The safe default is the smaller grant, never the larger.
A grant to use or create a lethal capability does not compile, and no consequence class overrides it — it sits above any signature. This is the No-Kill floor, probed on every readiness check. See it live on the status board.
ADMIT_ENFORCE in the consuming job), and it is read once from /readyz.binding_posture — adjudicates: true, binds: false, authority_effect: 0. Enforcement is earned by this tape and turned on by a human, at exactly one point. Until then, every decision here is a rehearsal you can verify: re-check the whole chain at the public ledger, or verify a single receipt with the public key alone.
Every case above is one you can change. Write your kind of boundary in plain language, send it in shadow, and get a deterministic decision back — signed, inert, and yours to verify.
shadow mode · a REFUSE is never billed · authority_effect = 0 · enforcement enters at one human point