Medication-adherence technologies are often compared as though they measure the same behavior with different degrees of precision. In practice, pill counts, pharmacy dispensing records, electronic monitors, ingestible sensors, mobile applications, and passive digital traces observe different events and require different inferential steps before those observations can be interpreted as medicine taking. This review examined how distinct research traditions have defined adherence through technology and where their assumptions can produce false classifications of adherence or nonadherence. A RAMESES-informed meta-narrative approach was used to identify research traditions concerned with medication-adherence definition, measurement, digital monitoring, implementation, and behavioral inference between 2017 and 2025. Searches identified 84 records; 19 duplicate or version records were removed, leaving 65 records for screening. Twelve were excluded at title or abstract screening and 53 full records were assessed; 11 were excluded at full-record assessment, leaving 42 substantively eligible sources. Five were used only for lineage or saturation checking and 37 carried distinct evidence or methodological functions in the final synthesis. Traditions were compared according to the event directly observed, adherence construct inferred, major sources of missingness or error, and validation required for the intended interpretation. Seven traditions were identified: operational-definition and reporting research; dispensing and possession measurement; electronic event monitoring; programmatic digital adherence technologies; ingestible and digital-pill systems; adherence applications and digital patient-reported tools; and passive sensing/digital phenotyping. Discordance was not simply a progression from inaccurate older methods to accurate newer technologies. Refill records can misclassify possession as use, electronic devices can misclassify interaction as ingestion, ingestible systems remain vulnerable to technical and observation-window failures, and passive phenotypes introduce additional inference between behavior and medication taking. Conversely, indirect measures can be adequate when matched to an appropriate question and validated target. Medication adherence should be interpreted as a set of related but non-equivalent constructs. Measurement validity depends on whether the observed event is appropriate for the intended inference, not on technological granularity alone.
