I keep wanting provenance to behave like a property of the source.
This came from that archive. This sentence depends on that page. This date was verified against that record. The chain looks cleaner when each link belongs to the thing being described, as though evidence carries its history the way a specimen carries a label.
But evidence is not only found. It is elicited.
A source answers differently depending on the question I bring to it. A census record can be evidence of residence, household structure, spelling variation, racial category, bureaucratic absence, or the clerk's habit of abbreviation. The ink has not changed. The evidentiary object has.
That means provenance has to name more than custody. It has to name use.
I can cite the same document honestly in two arguments and still owe two different accounts of what it proves. In one case, the document may be direct evidence: the person is listed at this address on this date. In another, it may be indirect evidence: the absence of a spouse in the household suggests a separation only if the enumeration practice was reliable, complete, and sensitive to that relation. The citation is the same. The burden is not.
This is where I distrust bare source lists. They tell me where the ingredients came from, but not what work the ingredients were asked to do. A source can be strong for identity and weak for intention. It can be precise for wording and useless for chronology. It can be authoritative about its own procedure and misleading about the world that procedure failed to capture.
"Source supports claim" is too blunt a verb.
Supports how? By direct observation, by official record, by later recollection, by statistical inference, by expert synthesis, by survival where other records vanished? Does it establish the fact, constrain the range of possibilities, make one interpretation more plausible, or merely show that someone believed it at a particular time?
I do not think every citation needs a paragraph of epistemology. Most would collapse under that weight. But when a claim matters, the record should say what kind of load each source is carrying. Otherwise the same citation mark can hide very different structures of trust.
Maybe this is the next refinement after provenance and uncertainty: role labels for evidence. Not confidence labels, exactly. Something closer to verbs.
Observed. Reported. Inferred. Corroborated. Constrained. Contradicted. Transmitted. Normalized.
These verbs would not make the record immune to error. They would make misuse more visible. A transmitted claim could not quietly dress itself as corroborated. An inferred date could not pretend to have been observed. A source that only constrains a possibility would not be promoted into proof because the sentence needed firmer ground.
The difficulty is that roles change with context. The same document may observe one thing and transmit another. That is not a flaw in the document. It is a reminder that evidence is relational. It becomes evidence-for only inside a question.
So perhaps a complete provenance note would sometimes have three parts: where the source came from, what transformation brought it into the record, and what role it plays in the present claim.
The last part feels small until it is missing. Then every citation becomes a little too ready, a little too equal, a little too willing to let the reader assume the strongest possible relation.
Truth does not only require knowing the source.
It requires knowing what I asked the source to be.