A measurement procedure should be tested under more than one notion of 'doing it again'.

Repeatability describes agreement when the same procedure is repeated under nearly unchanged conditions: same apparatus, operator, location and short time interval. It primarily probes short-term random variability.

Reproducibility asks whether compatible results are obtained after meaningful changes such as a different operator, instrument, laboratory, analysis implementation or time period. These changes expose systematic effects that repeated measurements in one configuration can miss.

Robustness checks deliberately vary analysis or experimental choices that should not materially change the physical conclusion. Examples include changing fit ranges within justified limits, reversing measurement order, using an alternative calibration, or analyzing independent subsets.

Suppose ten repeated voltage measurements agree within $0.1%$. That demonstrates strong repeatability. If replacing the voltmeter shifts the result by $2%$, the experiment is not yet reproducible at the claimed $0.1%$ level; the original repeated scatter understated the full measurement uncertainty.

Reproducibility does not require numerically identical results. Independent results should agree within their stated uncertainties and known differences in conditions. Disagreement is scientifically useful because it reveals missing uncertainty sources or incomplete models.

A reliable experiment therefore combines repeated data with deliberate variation of conditions. Precision established under one setup is evidence about repeatability; confidence in the physical result grows when independent routes lead to compatible conclusions.