Owned, version-controlled reference datasets used as competition index sources on crowdsource. These define what a competition asks participants to predict (the set of index keys) — kept here, under our control, rather than depending on third-party datasets that can go stale or disappear.

A daily GitHub Action (.github/workflows/refresh.yml) regenerates each snapshot from its upstream authority and commits only on change.

sp500/constituents.csv columns: Symbol, Security, GICS Sector, GICS Sub-Industry. Symbol is the index key (matched against Massive day-aggregate tickers at resolution).

bikeshare/citibike/stations.csv columns: station_id, name, lat, lon, capacity. station_id is the index key (matched against the Citi Bike station_status feed at resolution).

Competitions point their input_url at the raw file, e.g.:

https://raw.githubusercontent.com/1kbgz/crowdsource-datasets/main/sp500/constituents.csv

Add scripts/refresh_<name>.py (write <name>/<file>.csv, sorted, with a sanity floor on row count), wire it into the refresh workflow, and document it above.