Q-Psi After 564,840 Physical Quantum Shots
1. The Experimental Milestone
Over the course of 16 production quantum processor executions totaling 564,840 physical shots and 259.38 reported IBM runtime quantum seconds, Q-Psi investigated how compiler-synthesized state spaces behave when executed on real superconducting quantum hardware.
Rather than treating quantum advantage as a vague marketing narrative, we formulated strict pre-registered hypotheses across 14 discrete campaign families, locked all raw execution bitstrings with SHA-256 evidence seals, and executed an independent mathematical and statistical audit.
2. What Was Established (The Positive Evidence)
Under an opaque black-box verifier contract, Q-Psi demonstrated physical query reduction factors of 2.43x (N=4), 1.73x (N=8), and 1.19x (N=16) across 9/9 cases and 3/3 problem sizes. At N=16, the 95% confidence interval upper bound on quantum queries (7.360) remained strictly below classical expectation (8.5).
Physical single-shot execution across N=4..16 qubits demonstrated statistically significant query-complexity scaling advantage (α_Q = 0.1532 vs α_C = 0.6963, t = -30.65, p < 10-6).
Compiler liveness analysis compressed physical register width by a median 78.1% (mapping 16 logical qubits onto 2 physical slots) while satisfying pre-registered non-inferiority margins across 6/6 streaming circuits.
A 120-Qsec campaign sampled 65,138 candidate states from a 65,536-universe on IBM Heron, successfully producing 93 verified positive code repair records (1.593 MB, ~417k tokens) with 100% causal QPU provenance.
3. What Was NOT Established (The Strict Boundaries)
- No General Quantum Supremacy: Query reductions do not translate into end-to-end wall-clock speedup when compilation and QPU queuing times are included.
- No General Compiler Superiority: Q-Psi does not outperform industrial compilers Qiskit and TKET universally; benchmark competitiveness is workload-dependent.
- No Tested LLM Improvement: The Quantum Data Foundry demonstrated pipeline mechanics only; downstream LLM fine-tuning was not evaluated.
- Preserved Negative Comparators: In Calibration-Aware Layout, Qiskit O3 best-of-8 layout search outperformed Q-Psi layout (mean delta \(-0.0106\)). This falsifying evidence is preserved.