Product proof
That refusal is a product property, not a promise. Follow the question through policy, demonstration, evidence and limitation.
Buyer questions
Numeric benchmark results appear only after a protocol, environment, version, date, limitation and reproduction path exist.
Commercial policy is deterministic code, not a model's judgement: a renewal at or above the threshold stops and waits for a named human.
Benchmark dimension: Invalid-decision refusal
Limit: Proven for the built-in renewal object and its single value threshold. A general policy engine over arbitrary custom objects does not exist.
Inspect evidenceThe agent cannot approve on the human's behalf. A test asserts the refusal, so the boundary is a property of the system rather than a promise in a README.
Benchmark dimension: Human authority boundary
Limit: In local-development mode the actor is asserted, not authenticated: no authentication ships, so an actor header there is not an identity. This holds a boundary against an honest agent, not against an attacker with network access.
Inspect evidenceOne command tells an agent what an application actually is — packages, capabilities, resources, actions, policies, providers — read from checked-in source, in a single deterministic JSON report.
Benchmark dimension: Independent composition inspection
Limit: Source-only and read-only. It never opens the database, contacts a provider, reads a secret, or reports runtime, CI or authorization state — and it lists those blind spots as machine-readable limitations in its own output.
Inspect evidenceSQLite is Node's built-in node:sqlite. PostgreSQL requires one pinned runtime driver, pg@8.23.0. There is no ORM, no query builder, no build step and no framework underneath your framework.
Benchmark dimension: Source portability and hosted-runtime independence
Limit: Applications that select PostgreSQL carry pg@8.23.0. The SQLite path still needs no third-party driver. This is not a production-readiness claim and not shared-database tenancy; composition is dedicated-database, not row tenancy.
Inspect evidenceBenchmark architecture
A synthetic run is never presented as customer evidence. Where no equivalent baseline exists, comparisons cover architecture, responsibility and verification rather than invented competitor numbers.