Two checks, two different jobs
Suppose a team reports that a new search interface helps people find a document faster. Another analyst downloads its data, runs its code and obtains the same result. That is a useful check. The participants, however, are still the original participants. Nobody has yet shown that the result holds in another group.
The National Academies' 2019 report uses reproducibility for obtaining consistent computational results from the same inputs and analytical steps. Replicability concerns studies addressing the same question with new data. Terminology varies across disciplines, so a report should explain what was actually repeated.
Sources: s1
A matching answer can still contain a mistake
A calculation can reproduce perfectly while preserving a coding error. Conversely, a replication can disagree because the setting changed, the effect is variable or the original estimate was uncertain. One success does not settle a claim; one failure does not automatically dispose of it. The report recommends describing relevant uncertainty and the materials needed to repeat an analysis.
Sources: s1
Make both checks inspectable
A small website experiment could publish two separate records. One would identify the frozen event file, exclusions, code version and expected output. The other would describe a later observation window with new visits and the same predeclared outcome. The first would let readers inspect the arithmetic. A consistent second result would strengthen the case that the pattern extends beyond its first sample. Calling both records simply 'verified' would hide the distinction.