ARK-483 — Verification Decision Latency

ARK-483 — ExecutionProof Authorization Boundary Experiment — Verification Decision Latency Verdict: PASS (both implementations; worst-path p95 = 1.822 µs Python / 0.652 µs JavaScript — both ≤ preregistered 1000 µs ceiling) First RF latency measurement (Prospect Q2: what is the verification latency?). Component Under Test = the frozen ARK-458 guard, measured unchanged. Dual benches: V2 Python (imports the guard) + V1 JS (loads the guard's locked bytes via vm sandbox — no ARK-458 file modified) 3 decision paths (allow-exact / deny-first-dim / deny-last-dim) × 100,000 timed iterations, 5,000 warmup, cold-start recorded separately A verification decision costs ~1 µs on the reference machines — ~550×–1,500× under the preregistered ceiling Honest published finding: DENY is slower than ALLOW in both implementations — the mismatch branch formats a reason string, so the tail is dominated by explanation formatting, not comparison work Honest bounds: Measures the in-process decision floor only (excludes network, token retrieval, evidence persistence, process startup); bound to the reference machines in the results files. Working-example evidence only — does not legally validate any patent claim or certify RF-100 conformance. Published under the Remnant Fieldworks Standing Covenant. To God be the glory. Proof Before Power. Verification Before Execution.