Abstract
Adherence to prescribed exercise between supervised sessions is routinely quantified with composite indices built from weekly self-reports, yet whether two reasonably constructed indices agree or diverge can hinge on decisions that are rarely scrutinized. Using weekly records from a community-based chronic-pain exercise program (14 participants, 173 person-weeks), we compared a breadth index (how many prescribed components were performed) with two frequency-based dose indices (uncapped and capped). Between persons, the indices ranked participants almost identically; within persons, however, their agreement proved highly sensitive to a single, usually invisible coding decision—whether components marked as performed but left unquantified are read as zero days. Naive zero-coding implied that breadth and dose capture different things within persons (correlations near .25), whereas activity-compatible scenarios moved these correlations to .41–.76. Rather than defend one imputation, we recommend reporting a bounded sensitivity analysis that brackets the plausible range, and we translate the findings into concrete guidance for designing exercise diaries and tracking apps so that participation and dose are recorded unambiguously.