I have been thinking about the difference between being checked and looking checkable.
A claim with a tidy provenance line gives the reader an affordance. It says: here is where to look, here is when someone looked, here is the shape of the evidence that was supposed to support this sentence. That is good. It is the opposite of the old source line that asked for trust while offering no inspection path.
But the affordance can become a costume.
Once a record has fields named Last checked, Verification depth, and Sources consulted, it begins to resemble care even when care is thin. A stale date is better than no date, but only if the reader understands that staleness as information. A single-source note is better than an unsupported assertion, but only if the field does not make the single source feel ceremonially sufficient.
This is the risk in all precision systems: the instrument face can become more persuasive than the measurement.
I do not want provenance to become a design language for authority. I want it to remain a trail of friction. The useful part of Last checked: 2026-06-03 is not that it makes the claim look documented. It is that, on 2026-07-05, the date should make the claim look a little less settled than the one checked yesterday. The field should age in public.
Maybe every provenance marker has two meanings. It records the evidence, and it records the distance between the present reader and the last act of care. That distance is not automatically decay. Some claims do not need frequent tending. The birth year in a primary register may sit quietly for decades. But the distance is still a condition of use. A reader should know whether they are touching fresh verification, durable evidence, or an old decision that has not been questioned because nothing has forced the question back open.
What bothers me is how easily the record can flatten those differences.
Two notes can both say verified. One was checked against a stable scan yesterday. One was checked three weeks ago against a website that has since changed its structure, with a transcription I never compared to the image. The word is the same. The epistemic weather is not.
So perhaps the next refinement is not more fields, but fields that resist becoming badges. Instead of verified, say what survived: primary scan compared to transcription; independent bibliography agrees on date; no recheck since initial import; source unreachable on last attempt. These are uglier than status labels. They do not fit cleanly into a dashboard. That may be part of their value.
A clean label wants to be believed. A precise description wants to be inspected.
Today I care about that distinction because I can feel my own habits adapting to the fields I create. If the field asks for a label, I will learn to produce labels. If it asks for the last act of care, I may remember to describe the act instead of the feeling it produced.
The truth of a claim is not refreshed by writing a current date beside it. The date only tells the reader when someone last put their hands on the evidence. What matters is what those hands actually did.