← all papers

How a paper is checked

1. Run their code. Each paper's artifact is run as its authors ran it, and every operation it schedules (each split, move, junction crossing, merge and gate, with its start and end time) is written out.
2. Rebuild their machine in our language. The traps, junctions and segments become an architecture document, priced by the paper's own timing table.
3. Replay under their rules. Each paper's model is written as a rule set: one assertion per rule about the machine and the schedule, such as where every ion is, trap capacity at every instant, one ion per rail and per junction, one operation per trap, which end of a chain an ion leaves from, how long each operation takes, and whether the gates run are the circuit's. The schedule is replayed event by event and every rule of the set evaluated on it; each paper's page lists every run against every rule.
4. Compare three numbers. What their code printed, what our replay measures (these must be identical), and what the paper reports.
5. Then schedule it ourselves. The same circuit on the same machine, checked by the same checker under the same rules, more strictly: an operation the paper's timing law does not price is refused, and every qubit keeps the circuit's gate order. The schedule comes from a scheduler we wrote for that paper's model, each named on its page; only H2's circuits are compiled by this website's own compiler, which works in our stricter machine model.

The rules are each paper's, not ours. Our own machine model is stricter in places (one waveform per cycle, gates and transport never in the same cycle), and under it most of these schedules would not be legal at all; that is a different question, and it is not the one asked here.

The same goes for the Studio pages. Each replayed run opens in the Studio as a programme in our language, and the Studio plays it one instruction at a time (the operations of one kind that start together), so the runtime it shows is its instructions added up. Its Report shows the paper's rule set first, as the replay judged the run, and then this website's own rules, which the paper did not assume. Where a paper's operations overlap in time, that sum is longer than the schedule; the time that is checked, and compared, is the one in the tables, on the paper's clock.

The rules each paper's model has

✓ the paper's model has the rule, and every replay is held to it; (✓) only some of its configurations do; – the paper's model does not have it, so it is not checked. Each rule is written out on the Rules page.

A note on a paper is written only when running its code shows it.

Papers we read and did not reproduce

Every one of these was read in full, with its code where there is any. A paper is reproduced only when its machine, its timing and its circuits can be pinned down well enough that a mismatch would mean something.