Glossary
Point-in-time data
Point-in-time data records what a source said at each moment, so a model or eval can be restricted to information that existed on a given date.
For legal AI this prevents leakage: an eval question about the law on 1 March 2025 should not be answerable from an amendment published in June 2025. The history SKU makes this possible: version N is current from its fetchedAt until version N+1’s.
Arieta Data versions are observation snapshots, not reconstructions of the law in force; history starts when a document first entered the corpus and cannot be backfilled.
Sources
Related terms
Content hash
A content hash is a SHA-256 digest of a document’s normalised parsed text; Arieta Data records a new version whenever the hash changes.
Consolidated text
A consolidated text is a version of an act that merges the original with all its later amendments and corrigenda into one readable document; in most systems it is for information only, though some countries publish official consolidations.
Snapshot
An Arieta Data snapshot is a complete, immutable monthly cut of every SKU, identified YYYY-MM and shipped with a manifest of counts and SHA-256 checksums per shard.
Last updated · All terms