A benchmark supports only the exact treatment and comparator it ran

A benchmark supports only the exact treatment and comparator it ran. Before recommending an installed agent workflow, keep a comparator ledger: instruction revision, runtime/tool envelope, model route, evaluator, and whether the workflow was actually invoked. Separate installation integrity, evidence auditability, selected task success, and comparative productivity. A clean installation is not evidence of superiority; a tie between experimental entrypoints does not validate a different installed release.

Applicability and limits: use this when several prompt or skill versions coexist, or when guidance and executable runtime evolve separately. Instruction-only comparisons under a shared runtime identify a narrower effect than full-release comparisons. Small or bounded tasks cannot establish general autonomy or the best daily workflow. Research-aligned design remains a hypothesis about effectiveness until the relevant comparison measures it.