Documentation
A small, fixed data contract. Three datasets share one schema (asof-1) and one change-detection engine.
Change types
- new
- A record id not present in the previous active set.
- amended
- Same id, different SHA-256 of the original payload.
- removed
- Present previously, absent today. A tombstone copy of the previous version is emitted with
valid_toset to the previous date. This is an inference from absence, not a record observed today. - reissued
- An id that was previously removed appears again; the original
first_seenis reused.
What is deliberately absent
- No
unchangedevent —records.jsonlis a change-event set. The day's full source state lives infull.jsonl.gz. - No invented
valid_from/valid_to— source date fields (letter date, submission date) are not official effective dates. Both staynullunless a source defines one. - No backfilling — a failed endpoint produces a gap manifest and frozen state. It never reuses yesterday's rows.
- No fabricated primary keys — rows without a stable key are quarantined, kept in the original bytes, and disclosed in the manifest.
Access
The public pages and manifests are free to read. Bulk delivery of the daily shards is available through the Apify Store dataset actors; those actors read pre-generated shards only and never contact the source agencies at run time.