How data processing shapes cycling power and cadence indicators: Code and derived research data

Research reproducibility materials for an exploratory secondary analysis of public GoldenCheetah cycling telemetry. The principal cohort contains 890 duration-and-distance-corroborated rides from six athletes. The workflow evaluates whole-ride power-cadence correlations across seven active-pedalling thresholds and within-interval and between-interval slopes, correlations and covariance at 60, 180, 300 and 600 seconds. The release includes executable source acquisition and analysis code, pinned package versions, processing specifications, source provenance, variable definitions, computed ride-level and aggregate summaries, model diagnostics and publication figures. Source athlete UUIDs, recording filenames, timestamps, demographics and original sensor recordings are not redistributed. Release-local analysis labels preserve joins between computed tables; the full matching audit is regenerated locally from hash-verified provider files. Independent checks validate timestamp matching, corroboration arithmetic and estimator identities. The six-athlete sample limits population inference. Code is licensed MIT; original documentation, figures and derived summaries are licensed CC BY 4.0, to the extent rights apply. These licences do not apply to the third-party source dataset. This release is software and research supporting material, not a preregistration.