Error correction / Explorer
QEC Explorer
Real shots from the 2026 Willow chip, replayed round by round
Every round, the chip measures its checks. A check lights up when its result changes from the round before, which is the only trace an error leaves. At the end, the chip reads out the logical qubit, and each decoder's call on whether it flipped is marked right or wrong.
On every d3 validation shot in the release, 1,440,000 of them, Z-QEC Platinum is near level with AlphaQubit 2, 360,135 wrong calls against 355,369, Z-QEC Gold is just behind, and Z-QEC Gold run with no chip data makes the fewest of any decoder without chip data.
How to read it
- Blue and violet tiles are the X and Z checks. White dots are the data qubits that hold the logical qubit
- The bars under the patch count the checks that lit in each round
- Green circles are the final readout, compared against the last round of checks
- The ranked panel lists every decoder by its wrong calls on every validation shot at that distance
What you are watching
- This is a replay. The shots and every decoder's calls are stored, so everyone sees the same thing and nothing is decoded while you watch
- The shots are the first 100 validation shots of one run per code distance, from the 2026 Willow release
- The other decoders' calls are the ones published with the data. Ours are from offline decoding
- AlphaQubit 2 is a network trained on this chip's own data. Z-QEC was tested on the public release alone, with no access to the chip
The replay's own counts cover only the shots you have played, and on so few shots they move by chance; the ranked counts and the error rates in the decoder comparison are the figures to quote.
Data: Google Quantum AI, 2026 surface code release, Zenodo record 18896801, licensed CC BY 4.0. Changes: scored by us. Every name on this page is listed in the attributions.