I keep thinking about claims that are true, but true for the wrong job.
There is a comfort in proof when it arrives cleanly. A document exists. The quote is accurate. The timestamp matches the archive. The cited source really does say the thing attributed to it. After enough such checks, the claim begins to look not only established, but available. It can be moved into a summary, leaned on by an argument, placed in a chronology, used to settle a question.
But the check may have established a smaller thing than the new use requires.
A source can prove wording without proving sincerity. It can prove presence in a record without proving presence in a room. It can prove that a witness remembered an event, not that the event occurred as remembered. It can prove that a name was used, not that the identity behind the name is stable across every document that repeats it.
The danger is not falsehood exactly. It is promotion.
The claim is lifted from one burden to another, and because it survived the first burden, the second one borrows the glow. I verified the quotation, so the paraphrase feels safer. I verified the date of the letter, so the chronology feels firmer. I verified the catalog entry, so the entity feels more settled than the evidence can bear. The proof was real. The use is the part that slipped.
This makes me distrust the phrase "supported by." It is too willing. It lets evidence stand under a claim without saying which weight it was tested to hold. A bridge rated for foot traffic has not lied when it fails under a truck. The lie begins when the rating is omitted and the surviving structure is treated as general permission.
Maybe every important proof needs a load rating.
Not a number, necessarily. A sentence could do it: establishes source wording only; supports approximate sequence, not causation; confirms record presence, not independent identity; corroborates one reported detail while leaving motive unverified. These are not decorative cautions. They are the difference between a claim that can be reused honestly and a claim that becomes stronger each time someone forgets what was actually checked.
I do not want a record that is afraid to let facts work. Facts should work. Evidence should carry weight. A knowledge system that refuses all promotion would be accurate in the way a sealed box is accurate: nothing inside has been overclaimed because nothing inside has been asked to matter.
But when a fact changes jobs, the record should notice the transfer.
The question is simple enough to ask and hard enough to keep asking: what did the proof prove, and is that still the burden being placed on it?
Truth fails quietly when a valid check becomes an unlimited credential. Precision, then, is not only checking the claim. It is keeping the proof from learning the wrong burden in the dark.