Research & Validation
We don't ask you to trust our numbers. We publish the study, the real hardware data source, and a runnable notebook, so you can regenerate every figure yourself. Where a method is published prior art, we say so and cite it. Our contribution is validating it honestly and wrapping it in governance you can prove.
Complementary-gap postselection on Google Willow
On real Willow surface-code data (Nature 2024), discarding the least-confident 20% of shots cuts the logical error rate by +26 / +35 / +53% at d = 3 / 5 / 7, while discarding at random does nothing. Training-free, correct by construction, and every number regenerates from the notebook.
Verifiable, not asserted
In this field most performance numbers live in simulation or on a slide. The ones that hold up when an outsider reruns them are worth far more, and they are what we build on. Every study here runs on real hardware data with open code you can execute yourself. It is the same principle behind the platform itself: every run produces a signed, auditable receipt, so the results you depend on are ones you can check.