127 Quantum Seconds, 65,138 Candidate States and 93 Verified Coding Records
1. The Research Question
Can a physical quantum processor act as a causal combinatorial state selector to explore complex repository mutation universes, with classical executable test suites acting as automated ground-truth filters?
In our Data Foundry V2 campaign, we tested this end-to-end architecture on IBM Heron r2 (ibm_marrakesh).
2. The 120-Quantum-Second Campaign Numbers
3. The Classical Verification Throughput Bottleneck
A critical empirical discovery from this campaign is that candidate-state supply from the physical QPU exceeded classical verifier throughput.
The quantum processor sampled 65,138 candidate states in 127.58 seconds of quantum execution. However, compiling and running full regression suites in isolated sandboxes evaluated 158 candidates in 246.3 seconds. Thus, scaling synthetic code data generation is bounded not by quantum sampling throughput, but by the parallel execution throughput of classical compilers and test runners.
4. Strict Scientific Boundaries
The experiment demonstrates mechanical pipeline feasibility only. Downstream LLM fine-tuning, benchmark evaluation, and code repair accuracy comparisons were NOT EVALUATED. No claim of quantum training advantage or superior model quality is made.