Skip to content

Provenance

Every value in a Dataset names its Source and three times, so a buyer reads where a value came from and when it was true, and cites it.

Provenance is where a value came from and when it was true. Every value in a Dataset carries both, so a buyer can check it, cite it and reproduce it.

Three times

Each record carries three times, because each answers a different question.

FieldAnswers
observedAtWhen the fact was true
publishedAtWhen the Source published it
refreshedAtWhen the Dataset last read that Source

A rate read an hour ago from a bulletin published last month is a fresh read of an old fact. Buyers filter on the time that matters to them: a checkout on observedAt, an auditor on publishedAt.

The Source of every value

Every value names the Source it came from and, for a wrapped Source, the terms it was read under. A value two Sources disagreed on also keeps the losing observation, so the choice can be checked.

Reading the past

Nothing is overwritten. Every get takes at, a time or a Version name, and returns the record as it stood then. As-of reads covers reading a whole slice as it stood.

For agents

An agent that answers with a Dataset's value cites it: the value, its Source and its observedAt. The response carries all three, so the agent never has to guess where a number came from.