Provenance
Provenance is the record of where a piece of information came from, kept attached to that information so it can be checked later rather than trusted.
In a data system, provenance means every stored fact points at its origin: the email, the web page, the uploaded file, the message, the model run. Good provenance is openable. You should be able to click a job title and land on the email that said it.
It matters most where information is inferred rather than entered. When a model extracts twenty facts from a transcript, the facts look identical to ones a human typed. Without provenance there is no way to tell them apart afterwards, and no way to re-check the one that turned out to be wrong.
Provenance cannot be added later. A fact stored without its source is a fact whose source is gone, so it is a property of the write path or it does not exist at all.
Why it matters
AI systems make confident claims. A record that can be checked is worth repeating in a meeting; one that cannot be is a rumor with better formatting. Provenance is also what makes accuracy obligations under data protection law something you can demonstrate rather than assert.