Provenance
The recorded origin of a fact or output — what source it came from, and where.
category ▸ Data & Provenance
Provenance, in plain language.
Provenance is the documented lineage of information: for any fact an AI surfaces or any output it produces, provenance answers where it came from. At its finest grain — datapoint provenance — a single number on a report links back to the source document, page, and location it was drawn from.
Provenance is the antidote to the most dangerous AI failure mode: a confident, well-formatted answer with no basis in reality. When every figure carries its source, an ungrounded claim stands out instead of blending in, and a reviewer can verify rather than trust.
How Cortex implements it.
This term isn't abstract here — it maps to a real capability in the runtime. Here is exactly how Cortex enforces or relates to it.
Cortex pins each surfaced fact to its source through datapoint provenance — citations carry source, page, and bounding box, with an as-reported flag distinguishing a verbatim value from a derived one, and a value_hash proving the figure was not altered after extraction.
Ungrounded figures are flagged rather than silently asserted, and provenance threads into the broader lineage graph so a fact connects all the way back through the decision that used it.
Keep building the vocabulary.
These terms sit next to this one in the governed-AI model — follow the thread to see how the controls connect.
Trust Ledger
A tamper-evident, hash-chained record of every governed run, action, and approval.
Hash chain
A sequence of records where each one cryptographically includes the hash of the previous.
Tamper-evident
A record whose integrity can be proven — any change to it is detectable after the fact.
Ontology
The business object model — types, instances, and relationships — agents reason over.
See Provenance enforced, not just defined.
Book a walkthrough and watch the controls in this glossary return real verdicts, seal real evidence, and trace every fact back to its source.