What Would Settle the Leakage Attribution in Magic-State Cultivation on Willow

Bilar, Daniyel Yaacov · 2026-10-18 · publication/preprint · cc-by-4.0

Version of record (canonical): https://doi.org/10.5281/zenodo.23452623
Concept DOI (always latest PDF): https://doi.org/10.5281/zenodo.23452622
Download PDF (Zenodo): open
Companion software: 10.5281/zenodo.23452625 (catalog). Concept 10.5281/zenodo.23452624. The distance-3 simulator is not a result of the note.
Repository: github.com/chokmah-me/willow-follow-up-note
OSF: not opened
Sketchnote of draft v6. Panel A: experiment stem at 5.3 percent and simulation stem at 0.8 percent. Panel B: diamond at 6.6 inside the bar 6.2 to 7.1, solid line at 2 to close, dashed line at 3 for arm B. No arm has been run.

Sketchnote for v6 (CC BY 4.0). The results panel copies Figure 1. Full-size PNG.

Abstract

Google's 2026 magic-state cultivation experiment on Willow grafts a cultivated magic state from a distance-3 color code into a distance-5 code. In the grafted-code memory experiment, the logical error rate is about seven times a depolarizing simulation that omits leakage. The authors call leakage the likely main cause and ran no removal experiment. Variability in physical qubit performance is a separate route and does not by itself predict the leakage signatures. The public deposit has no files for the nine-cycle run. A pre-registered 2 by 2 has not been run. Until it runs, the attribution stays likely and not ablated. Companion software, which is not a result of the note: 10.5281/zenodo.23452625. Prefer concept DOI 10.5281/zenodo.23452622 for the always-latest PDF.

Keywords

Google Willow · magic state cultivation · leakage · grafted code · three-state readout · pre-registration

← All research