passengers_entrance, passengers_transported, and station_daily rebuilt through April 2026.
Vignettes
Added time-coverage-by-line charts and tables to the data dictionary for all four core datasets (previously only passengers_entrance had one).
Rewrote the “Core datasets” sections of the data dictionary to follow a single consistent structure (columns table, glimpse, time coverage).
Reordered the interchange-stations table by reference line and station name, and fixed assorted typos and wording issues.
metrosp 1.1.0
Datasets
Added calendar_spo: a São Paulo holiday and business-day calendar (2012–2030) covering national, state, and municipal holidays, for use in demand seasonality and business-day adjustments.
Removed the forecasts and forecast_accuracy exports. These were exported in 1.0.0 by mistake: metrosp is an observed-demand data package and forecasting belongs downstream, so they should never have shipped.
Removed the metro_lines export. Its line-name columns (line_name, line_name_pt) are already denormalized onto every passenger/station dataset, and the full line list (including planned and CPTM lines) is available in lines. It remains an internal join dimension in the ETL pipeline.
station_daily$line_number / station_daily$year are now integer (previously double), matching the other datasets. Values are unchanged.
Data quality
Station names are now canonicalized consistently across the demand datasets (station_averages, station_daily) and the geometry datasets (stations), so a station joins cleanly across sources.
Fixed footnote-digit and sponsor-name contamination in station names (e.g. stray trailing digits and parenthesized line numbers).
Trailing unpublished NA rows are now trimmed per line during assembly; interior NAs (e.g. station outages) are preserved. All datasets rebuilt.
Refreshed the 2017–2019 source CSV.
Fixed a duplicate row in stations (Vila Mariana, Line 1) caused by the GeoSampa import not deduplicating after the name/join cleanup step. Added a regression test asserting stations has no duplicate rows.
Fixed a row-skip bug in the passenger-by-line CSV parser (passengers_entrance, passengers_transported): a hardcoded per-year header offset was one row short of where the source’s “Jan” row actually landed, which shifted every month’s figures up one slot (February’s numbers were recorded as January’s, and so on) and mistook the annual total row for December. The offset is now detected dynamically per file instead of hardcoded, since the header length has drifted release to release.
Fixed a targets caching gap where the METRO CSV download target returned a constant directory path, so a fresh download never invalidated the downstream parsers (entrance_current, transported_current, averages_current, daily_current) – they kept serving stale cached data even right after a real re-download. The target is now content-hashed against the downloaded files themselves.
Data pipeline
The data-raw/ ETL now runs as a targets pipeline (targets::tar_make()), replacing the flag-driven run_pipeline.R orchestrator. Pipeline functions live in data-raw/R/; the legacy scripts remain and produce identical output. See CLAUDE.md for the workflow.
Gated refreshes (download, historical re-import, Dataverse) are controlled by environment-variable flags via tarchetypes::tar_force() and skip cleanly when the flags are off.
