Multi-Scale Validation of autoq-qec: Analytical Prediction, Simulation, and Real Hardware Execution on IBM Fez

This dataset and report validate the autoq-qec quantum error correction resource estimator (v3.4.3) across two circuit scales (4 and 85 logical qubits), using a quantum Pi estimator circuit executed on real IBM Fez hardware (Heron r2, 156 qubits). Across ten independent real-hardware trials (5 per scale), autoq-qec's fidelity predictions agreed with measured fidelity within 1 percentage point in every trial, though the safety margin shrank sharply with scale (~11x tighter at 85 qubits). A second, independent finding: simulating with a manually-specified, incomplete HardwareProfile underestimated real-hardware error by 23-105x, while simulating with the package's own fully-calibrated hardware profile (same physical chip) tracked real-hardware error 5-10x more closely -- traced to a UserWarning that autoq-qec already emits for incomplete profiles, silenced by a blanket warnings.filterwarnings('ignore') in the original test script. Includes the full validation report, both validation scripts (n=4 and n=85), and raw JSON results for all 20 runs (10 per scale: 5 simulation + 5 real hardware, with publicly verifiable IBM Quantum job IDs).