Transpiler Over-Optimization and Barrier Override on IBM ibm_torino Heron r1

"RETRACTED — March 2026: The conclusions in this paper are incorrect. Fidelity revival was likely due to coherent errors, not transpiler compression. Data remains valid but interpretation was wrong." EXPERIMENTAL OBSERVATION: IBM ibm_torino Transpiler Behavior This technical note documents two experimental observations on IBM ibm_torino quantum computer (Heron r1 architecture, 133 qubits), conducted March 8, 2026. OBSERVATION 1 - Transpiler Over-Optimization: Bell state circuit (H + CX) was tested with repeated CX gates (1x, 3x, 5x, 7x, 9x, 11x). Standard ZNE theory predicts fidelity should decrease with more gates. Instead, fidelity INCREASED from 85.2% (1x) to 91.2% (11x). IBM transpiler compressed all repeated gates into single CX operation. OBSERVATION 2 - Barrier Command Override: Barrier commands (designed to prevent transpiler optimization) were inserted between repeated CX gates. Barrier had NO effect - 3x-with-barrier circuit produced 91.3% fidelity, identical to 11x-without-barrier (91.2%). Transpiler ignored barrier instruction. CONCLUSION: Standard Zero Noise Extrapolation via OpenQASM is NOT possible on ibm_torino Heron r1 due to aggressive transpiler optimization. This behavior is specific to this architecture and has not been explicitly documented for Heron r1. All job IDs provided in attached PDF as verifiable proof.