Google Popular Times (GPT) measures how busy a place is relative to its peak; GTFS describes planned train movements. We examine how train frequency and nearby land use relate to station activity at six railway stations in Luxembourg City. We build three weekly layers per station: station GPT curves (SGP), weighted activity profiles of nearby points of interest (CPP), and hourly GTFS train frequencies (Freq). POIs within 500 m are cleaned and weighted by mean GPT, duty cycle, and spikiness. Because GPT is peak-normalized per location, series are standardized within each station before comparison; we use Pearson and partial correlations, a mismatch index m(t)=z(SGP)−z(Freq), and panel regressions with station fixed effects.

Activity co-moves with both catchment busyness (SGP–CPP r=0.39–0.85) and scheduled service (SGP–Freq r=0.73–0.90), and the latter persists after controlling for land use (partial r=0.39–0.76). Pooling the peak-normalized stations makes land use appear dominant, but this is a normalization artefact: with station fixed effects, frequency is the stronger within-station driver (β≈0.82) and land use secondary (β≈0.21), leading only at the dense hubs. Timing is well aligned everywhere; the gap is in service level. Supply–demand ratios span an order of magnitude (0.07–0.80), and mismatch concentrates on weekday evenings across all stations and on Kirchberg weekends, while suburban stops are over-supplied at weekends. Hollerich combines the largest suburban catchment with the weakest alignment and highest unmet demand-pressure. Future work will scale to the full network, exploit Luxembourg's fare-free transit as a natural experiment, and add real-time counts.