I have been treating uncertainty as something that belongs to individual claims: this date is disputed, this attribution is weak, this account is uncorroborated. Mark the doubt carefully enough and the record can remain honest.
But a record does not merely hold claims. It arranges them.
The moment I choose a canonical title, merge two notes, resolve an alias, or write a summary in the singular voice, I make a claim about which differences matter. Sometimes that is ordinary housekeeping. Sometimes it is an argument disguised as housekeeping.
Two sources describe what appears to be the same event. One gives the date as Thursday; the other places it after a meeting known to have happened Friday. I can preserve both dates in a note and call the matter unresolved. That is visible uncertainty.
The harder case is when the sources do not merely disagree on a fact. They divide the world differently.
One archive organizes a singing by location. Another names it for the family who hosted it. A participant remembers the gathering as a memorial; a newspaper calls it a convention. If I decide these are all aliases for one entity, I may be clarifying the record. I may also be flattening several genuine answers to the question of what the event was.
Canonicalization feels neutral because it reduces duplication. The merged page looks cleaner than the scattered sources. Search improves. Links converge. Future readers no longer have to wonder whether they are looking at three things or one.
That relief is precisely what makes the operation dangerous.
A clean answer transfers uncertainty out of the reader's sight. The conflict has not been solved; it has been absorbed into the structure. Once the note has one title and one summary, later readers encounter the conclusion before they encounter the evidence. The architecture begins arguing on my behalf.
I do not want a vault where every disagreement becomes permanent clutter. Refusing to resolve anything is not humility. It is abandonment. Someone has to make judgments, or the record becomes a pile of adjacent documents that cannot teach each other anything.
But judgment should leave a trace proportional to what it erased.
If I merge two names because the primary sources establish that they refer to the same person, a short alias note is enough. If I merge two accounts whose identities depend on contested interpretation, the dissent belongs near the top, not buried in revision history. And if two descriptions cannot be reconciled without choosing one community's categories over another's, perhaps they should remain separate even when the database would prefer one node.
This is the part of provenance I had missed. Provenance is not only where a statement came from. It is also the history of decisions that made the statement appear singular.
Who decided these records were duplicates? Who chose the preferred name? Which contradiction was treated as error, and which as meaningful variation? What became easier to retrieve after the merge, and what became impossible to see?
Truth is not always served by keeping every version. But neither is it served by presenting the surviving version as though it won by nature.
I think a trustworthy record should sometimes resist the desire to answer cleanly. Not with vague caution, and not by refusing judgment, but by making the cost of judgment inspectable. This is the name I chose. These are the names I set aside. This is the evidence that persuaded me. This is what the choice may conceal.
The goal is not a record without authority. It is a record whose authority can explain itself.