ExecutionProof Enterprise Agent Boundary Testbed — ARK-493 through ARK-498 (Frozen Preregistration v1.1)

This deposition contains the frozen preregistration (v1.1), the complete testbed source, and the full evidence artifacts for six preregistered experiments (ARK-493 through ARK-498) characterizing the enforcement boundary of the ExecutionProof agent-authorization gate. The gate verifies six dimensions (actor identity, live authority, evidence, policy version, system state, exact-action integrity) and emits ALLOW / DENY / HOLD, recording every decision in a signed, hash-chained, independently reconstructable ProofRecord. Results (single reproducible run, preregistration hash verified at execution): 161 scored cases, 161 PASS, 0 FAIL, 0 enforcement leaks; dual-guard (in-process + isolated-subprocess) agreement on all 161 records. ARK-497 demonstrates independent reconstruction of all nine decision elements and 10/10 single-field tamper detection with zero false positives by a verifier statically proven to import nothing from the application. ARK-498 characterizes production-like overhead behind a real HTTP/loopback-TCP boundary (~1,810 requests) and meets all six frozen hard criteria: fail-closed leak count 0, zero duplicate executions, 100% ProofRecord completeness, clean error accounting, recovery with no automatic re-execution of denied requests, and 100% independent signature verification. Latency/throughput data are explicitly published as production-like overhead characterization, not a benchmark certification and not a production SLA, and must not be compared to the prior in-process microsecond testbed (ARK-483–492). Standing reconciled corpus across all series to date: 72 preregistered experiments, 232 case records, 229 PASS, 2 FAIL, 1 GATE-STOP, across 9 repositories. Negative results (FAIL, GATE-STOP) are retained unchanged.