ghlake

Every public GitHub event from GH Archive, loaded hourly by github.com/aishitdua/ghlake.

• silver/dt=YYYY-MM-DD/hour=H/events.parquet: one row per event, deduplicated on event_id.
• gold/*.parquet: daily and hourly aggregates. gold/stats.json has the running totals.

catalog.duckdb holds views (silver.events, gold.*) over the Parquet files, rebuilt every run:

attach 'hf://datasets/aishitdua/ghlake/catalog.duckdb' as ghlake;
select repo_name, stars_7d from ghlake.gold.star_velocity limit 10;
-- filter silver on dt (and hour) so DuckDB reads only those files
select event_type, count(*) from ghlake.silver.events where dt = '2026-09-25' group by 1;