A quantum computing lab rubric should reward prediction, reasoning, revision, and evidence.

A learner can run copied code and receive plausible counts without understanding gate order, bit mapping, sampling, source semantics, backend choice, or limitations. Conversely, a failed run can reveal strong understanding when the learner identifies the fault, tests a hypothesis, and records a justified revision.

Score the evidence trail. The rubric should reward what the learner predicted, built, inspected, executed, inferred, questioned, and improved—not the prestige of the provider or the cosmetic quality of one chart.

A practical starting model totals 100 points: prediction 10; circuit and source correctness 20; execution record 15; result interpretation 20; limitations 10; revision 10; evidence and provenance 15. Institutions should adapt weights to the learning outcomes, accessibility needs, assignment level, and academic policy.

Publish descriptors at four performance levels—exemplary, proficient, developing, and insufficient—before the lab. A point total without descriptors invites inconsistent grading and gives learners little guidance about how to improve.

Prediction earns credit when the learner states the expected qualitative or quantitative behavior, assumptions, and a falsifying observation before execution. It should be specific enough to compare with the result without demanding that finite-shot output match an exact distribution.

Circuit and source correctness covers gate sequence, controls, parameters, qubit and classical-bit mapping, measurement placement, and an explanation of how the visual workflow corresponds to generated Qiskit, Cirq, or OpenQASM where used. Correct code without an explanation is not full credit.

Execution evidence should identify workflow and source snapshot, simulator or provider route, non-secret configuration, shots or precision, timestamps, result, errors or retries, and any transformation applied after the raw output. Credentials, billing records, private emails, and hidden administration notes do not belong in a shared submission.

Interpretation compares the result with the prediction or baseline, explains sampling and noise where relevant, distinguishes observation from inference, and avoids an advantage claim from a classroom-scale demonstration. A polished chart with no reasoning receives limited credit.

Limitations can include finite shots, simulator assumptions, mapping overhead, device noise, stale calibration context, small problem size, missing baseline, unsupported operations, or uncertainty about an interpretation. The strongest response connects each limitation to the conclusion it constrains.

Revision asks the learner to act on feedback or evidence: correct the source, change a test, add a baseline, rerun with a justified setting, or explain why another run would not be useful. Revision credit should reflect the quality of the reasoning, not simply whether the second result looks better.

Before marking a cohort, instructors and teaching assistants should score a small shared sample, compare disagreements, clarify descriptors, and document accepted variations. Repeat moderation when the lab changes from simulator-only work to provider context or from foundation circuits to hybrid algorithms.

Use anonymized exemplars where policy permits. Show one strong explanation, one technically correct but under-explained submission, and one responsibly diagnosed failure so learners understand that the rubric values evidence rather than a predetermined output.

QFlow Academy can connect lesson state, workflow, generated source, run checks, attempts, feedback, progress, and reviewer-safe evidence. Teachers use the same learner experience as other members; their additional documented permission concerns academic-source review, and the review interface itself remains admin-gated. Institutional teaching guidance, grading, and appeals stay outside that product-role claim.

QFlow does not determine academic credit, accommodations, misconduct findings, appeals, professional competence, accreditation, or provider certification. IBM Quantum Learning and Qiskit resources remain independent learning materials; their use does not create IBM endorsement or accreditation of the rubric.