Part 6 · 24-qubit hardware
The same logical 24-qubit QASM and observable were executed through a direct IBM route and through Fire Opal. The Fire Opal value was substantially closer to the exact reference, but the runs were not simultaneous.
The numbers
The exact central imbalance is 0.9872906000. Direct IBM job d9rh6inpemts73csjq80 on ibm_kingston used 4096 shots and returned 0.448446. Fire Opal returned 0.751254. After postselection onto exactly twelve particles, the corresponding values were 0.496061 and 0.864706.
| Route | Imbalance | Absolute error | Fixed-N weight |
|---|---|---|---|
| Exact | 0.987291 | 0 | 1 |
| Direct IBM | 0.448446 | 0.538845 | 0.1653 |
| Fire Opal | 0.751254 | 0.236037 | 0.2778 |
| Direct IBM, fixed-N | 0.496061 | 0.491230 | postselected |
| Fire Opal, fixed-N | 0.864706 | 0.122585 | postselected |
In this comparison the raw absolute error is 0.302808 smaller through Fire Opal; after fixed-particle-number postselection the reduction is 0.368645. This is a strong practical result for this single run and observable.
Why this is not yet causal A/B evidence
Quantum hardware changes continuously through calibration and drift. The two routes were not executed simultaneously. Logical circuit, backend request, shots and observable were matched, but the physical noise realisation was not. We therefore say “Fire Opal was closer to exact in this comparison”, not “Fire Opal always causes this improvement”.
What postselection adds
The model conserves particle number, so outcomes outside twelve particles are invalid within it. Postselection removes those shots, but retains only 16.5% of the direct IBM weight and 27.8% of the Fire Opal weight. The improved central value must be reported together with that retention fraction.
Sources and reproducibility
Project status: 8 August 2026. Numerical values come from the frozen local research artifacts; the original claim boundaries are deliberately preserved.