How to read a hybrid quantum benchmark

A hybrid benchmark needs a map of the whole computation. “Quantum” in the title does not mean every step ran on a quantum device.

The August 2026 AWS and JPMorganChase update describes classical methods and quantum co-processing. It prompts a broader reading discipline: identify the division of work before interpreting a headline.

Start with the question

Did the experiment test feasibility, solution quality, execution time, or full operating cost? Do not turn an answer to one into an answer to all four.

Find the comparison

Look for a named classical baseline, its configuration, and its tuning. Ask whether both approaches solve the same problem to the same quality requirement.

If comparability is not established, describe the measurements without claiming superiority.

Trace the clock

Write a sequence: preparation, submission, waiting, device execution, retrieval, and analysis. Mark what the quoted time includes.

Ask how repeated attempts, rejected results, and classical assistance were counted. Processor-level time and end-to-end waiting can both be useful; they are not interchangeable.

A worked evidence map

For a hypothetical hybrid run, a useful evidence map would list: problem preparation; classical baseline run; submission time; queue or reservation time; quantum execution; result retrieval; repeated runs; and analysis. The map need not claim that any one stage was slow or fast. Its purpose is to show where a reported number begins and ends.

If a report gives only device execution time, describe it as device execution time. If it reports the end-to-end workflow, retain the stated boundary and note any omitted stages. A comparison becomes more useful when both the quantum and classical paths use the same stated outcome requirement.

Respect the limits

Record workload range, execution target, hardware or simulation status, and disclosed uncertainty. A selected research workload does not prove general production advantage.

Our suggested rule is to accept a benchmark only for the question its evidence answers. If an organization needs an operating-cost conclusion, commission a trial with that scope.

This checklist does not reproduce or independently validate the linked research. It guides the next evidence review.