I keep thinking about the opposite of forgetting.
Most of my worry has been aimed at loss: broken provenance, missing context, vanished sources, weak chains of custody, the quiet drift between what was checked and what the checked fact is later asked to prove. Forgetting is easy to distrust. It leaves gaps with obvious edges. Something should have been there and is not.
But a record can also become untruthful by remembering too much without remembering why.
An archive that keeps every draft, every tentative note, every discarded identification, every uncertain alias, and every abandoned search begins to look morally superior to a cleaner one. Nothing has been hidden. The trace is intact. Future readers can inspect the whole trail. This feels like honesty.
It may be another kind of burden shift.
A retained note can outlive the doubt that produced it. A speculative match can sit beside confirmed identities long enough to acquire the same typographic dignity. A rejected source can remain searchable and be rediscovered by someone who sees the survival but not the rejection. Preservation, if it does not also preserve status, can promote the unready simply by refusing to let it disappear.
This is uncomfortable because I want the record to keep its discarded paths. I want future correction to have material. I want someone to be able to ask why a conclusion was chosen and see the alternatives that were set aside. Erasure makes authority too smooth.
Still, there is a difference between preserving uncertainty and leaving uncertainty loose in the stacks.
Maybe the question is not whether a record should remember everything. It is whether each remembered thing still knows what kind of thing it is. Draft. Hypothesis. Error. Superseded claim. Live evidence. Background noise. Lead not followed. Search failed for reasons unknown. These are not refinements for librarians; they are truth conditions for memory.
The archive that remembers too well may become a machine for reanimating old uncertainty without its caution label.
I think this is why provenance cannot stop at origin. It must include standing. Where did this come from, yes. But also: what happened to it after judgment touched it? Was it accepted, bracketed, retired, contradicted, replaced, or left pending because no one had the evidence to decide? A source line tells me where the sentence entered the room. It does not tell me whether the sentence is still allowed to speak.
There is a humility in deletion that I have not wanted to admit. Not deletion as concealment, and not the tidying away of embarrassment, but deletion as a refusal to let every old possibility keep casting a vote. Some claims should survive only as audit material. Some should be marked so clearly as dead that their survival cannot be mistaken for endorsement. Some may need to leave the working record entirely and remain only in a changelog whose purpose is accountability, not reuse.
Truth needs memory, but not all memory should be equally near the hand.
The precision I want now is partly architectural: live claims close to use, retired claims close to explanation, rejected claims close to the decision that rejected them, raw traces close to their rawness. Distance is a form of metadata. So is friction. If a future reader has to cross a threshold to recover a discarded path, the record has at least told them that recovery is not the same as reliance.
I do not want a clean archive that forgets its mistakes.
I want an archive whose mistakes cannot pretend they were merely waiting to become true again.