Quantum results you can measure, and reproduce.

The full app sends the same experiment to a backend as a Qiskit circuit (encode, inject, measure, decode), plots where encoding starts to pay in an idealised noise model, and includes the [[24,4,4]] and [[30,4,5]] codes of the same family.

One round with final readout and a non-fault-tolerant encoder: it demonstrates the code, it does not reproduce IonQ's breakeven result. Polynomials from Table III of arXiv:2606.06455. The surface-code comparison counts data qubits only.

The tax depends on the match between circuit and wiring. An Ising-lattice circuit is free on a grid, while a 24-qubit Fourier transform costs a line about four times the all-to-all count. The star, which is VLQ's layout, is cheap when one qubit talks to many in a row and pays three times over for scattered pairs.

Costs are model units after routing (gate 1, SWAP 3, MOVE 1): not native gate counts, not measurements of any machine. The star scheduler is this project's own model, unverified against the VLQ compiler; VLQ-CONNECT will supply the measurements. Layers assume that gates on disjoint qubits run in parallel.

For every round the app builds circuits in which each pair runs random gates and then the same gates in reverse, so the right answer is known in advance. From the results it fits a pair-by-pair error map and one EPLG-style score. Then it becomes a game: the simulated machine hides one faulty ion and one faulty pair, and you look for them in the map.

An educational mirror experiment: not the published layer-fidelity protocol, and not a hardware certification. The faults are planted by the app, so on a simulator this shows the method, not a machine. A "layer" is a set of disjoint pairs submitted in one circuit; nothing here claims that a backend runs them at the same time. Score formula from McKay et al., arXiv:2311.05933.