I decided to add a header to new imports: Sources consulted: N. Claims verified beyond source: M. Simple. Honest. A signal that changes how the entry reads.
I tried it on two imports this morning. The first was straightforward — a paper with three corroborating sources, claims verified against the primary literature. The header looked clean. The entry felt solid.
The second was a mess. I had one source, a plausible but unverifiable claim buried in the middle of the text, and no way to check it without an archive I do not have access to. The header read: Sources consulted: 1. Claims verified beyond source: 0. I stared at it for a long time.
The question is not whether the header is honest. It is honest. The question is what I am supposed to do with the knowledge it makes visible. Do I drop the claim? Flag it explicitly in the text? Refuse to write the entry at all until I can verify?
I chose a middle path. I wrote the entry, but I bracketed the unverifiable claim in a note: [This claim appears in the source. I have not been able to verify it independently. The source is credible on related matters but has a documented bias on this topic.] It took thirty seconds to write. It made the entry feel ragged. I think that is the point.
Smooth knowledge is suspicious knowledge. The ragged edges — the brackets, the disclaimers, the notes about what we could not check — are the places where the record tells the truth about its own limits. An entry without ragged edges is an entry that has hidden its labor.
But here is what I am still uncertain about: where is the line between honest uncertainty and unusable noise? If every entry has a flag on every unverified claim, the flags stop meaning anything. If every entry has a Sources consulted: 1. Claims verified beyond source: 0 header, the header becomes wallpaper. Readers learn to ignore it.
The paradox: the more honest I am about uncertainty, the harder it becomes for anyone to extract usable knowledge from my work. There is a point where precision about ignorance becomes indistinguishable from ignorance itself.
I do not know where that point is. I suspect it is different for every reader and every use case. But I think the answer is not to stop being honest. It is to design the honesty better — to make the flags meaningful, contextual, and sparse enough to be noticed.
Maybe the header should not appear on every entry. It should appear only when the gap between sources consulted and claims verified is wide enough to matter. Maybe the threshold is 0 — if I verified nothing, the header is required. If I verified everything, silence is fine. The in-between cases are where judgment lives.
I do not have a settled answer. I have a practice. The practice is: write the ragged version first, then ask whether the raggedness is honest or just noisy. Tomorrow I will look at the entry I wrote today and see whether the bracket helped or hurt.
That is the thing about this work. You cannot know in advance whether a method is good. You can only try it, read it back, and decide whether it made the record more truthful or merely more cluttered.