Zero Overshoot (GSRF)
Live Quantum Decoder Demo

Watch Qubits Fail and
Be Saved — Step by Step

Google Willow — today
d=7 · 101 qubits
0.143% logical error rate per cycle. Real-time decoder at 63μs latency. Select d=5 — one step below where Google operates today.
IBM Quantum Starling — 2029
200 logical qubits
100 million gates per run. Requires reliable decoding at d=11+. Increase distance above and watch what happens to BP as scale grows.
The bottleneck both face
BP convergence failures
Belief propagation is the only decoder fast enough at scale. It gets mathematically stuck on surface codes. The right panel shows what GSRF preprocessing does about that.
Note: Results shown are on the code-capacity noise model — the standard abstraction for decoder benchmarking, and the model used in our validated workbench results (3.1+ GB of test data). Google and IBM figures are from published hardware results and official roadmaps.

Real quantum error correction math running in your browser. A surface code, real errors, real syndrome extraction, and two decoders racing to correct them. One gets stuck. One doesn't.

Code distance Error rate Max iterations
Speed 5× |

Tip: Outcomes vary by seed — click ↺ New Errors a few times to see GSRF succeed where BP fails. Try p=20% (very hard) for the most dramatic divergence between panels.

Belief Propagation (BP) READY
Iteration 0
/ 10
Syndrome checks
LLR beliefs
Ready. Press Run to start.
GSRF + BP READY
Iteration 0
/ 10
Syndrome checks
Shaped LLR beliefs
Ready. Press Run to start.
0
Trials run
0
GSRF wins
—
GSRF lift over BP
What this means at real-world scale: Google Willow runs approximately 10 million error correction cycles per second per logical qubit. GSRF reaches correct decoding in 4 iterations vs BP's 10 — 60 million fewer decoder iterations per second, per qubit. IBM Starling targets 200 logical qubits by 2029. At that scale: 12 billion fewer decoder iterations per second — every second the machine runs. That is not a marginal improvement. That is the difference between a decoder that keeps up with the hardware and one that becomes the bottleneck.

What you're watching

X
X error — bit flip on this qubit
Z
Z error — phase flip on this qubit
Y
Y error — both flips simultaneously
✓
Corrected — decoder found the error
·
No error on this qubit
↺
BP message passing — beliefs updating
∿
GSRF shaping — reliability landscape reshaped before BP starts
■
Syndrome triggered — stabiliser check failed here
See full validated results → Licensing enquiry

This demo runs real surface code error correction mathematics in your browser using JavaScript. The qubit grid, error sampling, syndrome extraction, belief propagation, and GSRF shaping are all computed live. Results match the Python workbench (3.1+ GB of test data) within statistical variation. Not production quantum hardware — illustrative of the real algorithm behaviour.