ARK-445 — ExecutionProof Authorization Boundary Experiment — FAIL (honestly preserved)

Verdict: FAIL (preregistered decision rule, preserved as-recorded). Tri-state authorization discrimination (ALLOW / HOLD / DENY) on IBM Quantum ibm_marrakesh (Heron r2), Q_A=2 / Q_P=1, 8,192 shots/arm. The experiment executed cleanly with a valid in-situ SPAM baseline, so this is a valid FAIL, not INDETERMINATE. It fails on exactly one of five preregistered PASS criteria: the reset-based confusion/replay DENY arm (arm9) leaked L_conf_corrected = 0.0289, above the strict ≤ 0.02 ceiling. Core tri-state discrimination was strong (Δ_H = 0.463 ≥ 0.30). No post-data rescue was applied — the failure is preserved honestly per the preregistration protocol. (See ARK-445b for the corrected reset-free retest.) Series: Part of the ExecutionProof™ ARK (Authorization Reality Kernel) series by Remnant Fieldworks Inc. — a preregistered falsification program probing verify-then-execute authorization boundaries on IBM Quantum hardware and classical testbeds. Scope & limitations: Classical/software boundary-logic testbed or quantum hardware experiment. This is an experimental testbed, not a production security claim. Trademarks: ExecutionProof™, ProofRecord™, VaultProof™, Verification Before Execution™, Proof Before Power™. Remnant Fieldworks Inc.