Measurement
Sixty-two logical qubits, certified in one job.
Thirty-one independent encoded pairs — 62 logical qubits — each certified above its witness bound with its own confidence interval, in a single submission.
Key takeaways
- Thirty-one [[4,2,2]] encoded pairs — 62 logical qubits — were placed across a Heron-class processor and certified in one submission.
- Every block cleared the 0.5 entanglement-witness bound with its full confidence interval: 31 of 31, mean witness 0.9964, weakest block 0.9790.
- Block failures are statistically independent — the joint pass rate matches the independent-blocks model to about 1%.
- The scaling boundary is measured, not hidden: joint acceptance across all 31 blocks is 9.4% (Z) / 6.0% (X) — post-selection does not scale, which is why the roadmap is built on correction.
What counts as a logical qubit here?
Four physical qubits encode two logical qubits in the [[4,2,2]] error-detecting code — a textbook construction, published on several platforms before us. Distance 2 means errors are detected and flagged, not corrected; shots that fail the check are discarded before scoring, and the acceptance rate is published beside every number. And this is a fleet of thirty-one independent pairs — not one 62-qubit entangled state. Each pair is certified on its own.
What did we measure?
One job. Thirty-one blocks, placed across the processor with topology-aware placement, each measuring the entanglement witness of its own encoded pair. Mean per-block acceptance after the detection check: 92.1%.
What did the numbers show?
All thirty-one blocks came out above the 0.5 witness bound, each with its full 95% confidence interval clearing it. The mean witness is 0.9964; the weakest block reads 0.9790. The blocks also fail independently: the measured joint pass rate matches the product of the individual rates to about 1%, so one certification does not lean on another.
What doesn’t scale — and why that matters?
Post-selection multiplies. Accepting ~92% of shots per block means accepting 9.4% (Z) / 6.0% (X) jointly across all thirty-one — a boundary we measured and publish, because it is the cleanest demonstration of why detection is not correction at scale. Crossing that boundary takes error correction, and that is the requirement our processor architecture is designed around.
How can you check it?
The raw per-shot arrays were archived at submission and every number re-derives from them offline; a re-decode on 2026-08-15 reproduced the certification. The job ID and full conditions are on the ledger.
Conditions. [[4,2,2]] error-detecting code (distance 2 — encoded and detected, not error correction), post-selected: mean per-block acceptance 92.1%, joint acceptance across all 31 blocks 9.4% (Z) / 6.0% (X); Heron-class processor; raw per-shot arrays archived and re-decoded 2026-08-15. Job ID d9mf6308csec73fa3o7g. Full conditions: the ledger.