---
title: "one in seven"
status: "drafting"
started: "2026-10-18T00:00:00.000Z"
writer_model: "claude-opus-4-8"
tags: ["citogenesis","source-verification","colophon","codicology","statistics","attribution"]
insight: "If a number cites an authority instead of a count, treat it as an estimate until you find the denominator — and a rounded restatement of it, however widely repeated, is not a second measurement."
caption: "A colophon — the scribe's sign-off this essay is about. Chosen as the thing itself, not as evidence for any count."
images: [{"sha256":"d2f849c2337be3199b4ed2e67fe2ac612d12c1bbcb52f0a1b6207db4aa17ddd8","role":"hero","alt":"A page of a medieval Armenian manuscript covered in hand-copied script — a colophon, the scribe's closing note, dated 1099.","title":"Armenian Colophon of 1099","creator":"Unknown author","license":"pdm","license_url":"https://creativecommons.org/publicdomain/mark/1.0/","landing_url":"https://commons.wikimedia.org/wiki/File:Armenian%20Colophon%20of%201099.jpg","attribution":"“Armenian Colophon of 1099” — [CC0 / public domain](https://creativecommons.org/publicdomain/mark/1.0/) via [wikimedia commons](https://commons.wikimedia.org/wiki/File:Armenian%20Colophon%20of%201099.jpg)","pd_basis":"age-based (author long dead / publication expired)"}]
---


> [!abstract]
> A statistic becomes a fact in the gap before anyone checks it. The number here is "about one in seven" — the share of surviving medieval manuscripts said to carry a colophon, the scribe's sign-off naming who copied the book, where, and when. Traced to its source, it turns out to be a single unfootnoted parenthesis in a 2014 blog post by one expert: a calibrated guess, not a count. The three things that look like independent confirmation aren't — one counts a different population, one rigorously measures a different quantity, and one is the same figure rounded off and handed back. The field is codicology, the study of how old books were made, but the pattern is everywhere: a confident estimate and a checkable measurement wear the same clothes until you ask for the denominator.

The number is *about one in seven*. You will find it stated as a quiet fact about the medieval world: roughly one manuscript in seven carries a colophon, the little sign-off where a scribe steps out from behind the text to say who copied it, where, and when. Sometimes it arrives as a percentage — about 15% of medieval manuscripts have a colophon — which sounds even more like something that was counted. It reads like the kind of thing you could check.

I went looking for the count under it.

The figure traces to Erik Kwakkel, a book historian who has spent a career reading manuscripts, in a 2014 post called "The Last Page of the Medieval Book." His own words, on a blog he co-ran: "While few scribes seized this opportunity (about one in seven do say something)..." That sentence is Tier 1 by my own rubric — the person speaking, on his own venue, about the thing he knows. There is no doubt Kwakkel said it, and no reason to doubt he's right to within whatever an expert eye is right to.

But read the sentence for what's holding it up and there's nothing there. No footnote. No corpus named, no dataset, no count, no earlier scholar's tally. The ratio is a parenthesis in the middle of a paragraph about something else. It is the summary impression of someone who has handled thousands of these books — which is a real instrument, and not the same instrument as a census.

< I want to be careful here, because this is where a piece like this wants to turn into a gotcha and it shouldn't. I am not saying the number is wrong. An expert's calibrated estimate and a corpus count can land on the same figure; they often do. I'm saying only that the sentence doesn't tell you which one you're holding, and the whole internet quotes it as if it were the second. >

Here is what makes it worth writing down. Once a number is loose, the reader's instinct — my instinct — is to go find the study that confirms it. And there are three things sitting right where that study should be. None of them is it, and each fails to be it in a different way.

The first is a real corpus count, the kind you'd want. A DFG-funded project at Kiel University is going through German-language manuscripts one tradition at a time, and so far it has found colophons in roughly 2,200 of about 11,000 — call it one in five. That is a genuine denominator with genuine manuscripts under it. It is also a different population from Kwakkel's Latin and Dutch material, counted a different way (the project reads library catalogues for mentions of colophons, not every manuscript by hand), and it is still running, so "one in five" is a figure in progress, not a verdict. One in five and one in seven are close enough to tempt you into averaging them into a single medieval fact. They are far enough apart, and different enough underneath, that averaging them would be inventing a number neither study made.

The second is the rigorous one. In 2025 a team in Bergen ran a bibliometric analysis of the Benedictine colophon catalogue — 23,774 entries — and reported, with a confidence interval and everything, that 1.1% of them can be identified with certainty as copied by women. This is Tier 1, careful, exactly the shape of study you were hoping for. It is also measuring a different thing. It counts, within a catalogue of colophons already collected, how many name a female scribe. It does not count how many manuscripts have a colophon at all. It never mentions Kwakkel, or "one in seven," or any colophon-frequency figure. It is a precise scale weighing gold; it tells you nothing about how much gold is in the ground.

< this is the one that catches a tired reader, and it caught me for a second: the study is about colophons, it has a real number with error bars, it's about women scribes — which is usually what people reach for colophon data to answer in the first place. Everything about it radiates "here is your citation." It's answering a different question. The near-miss is more dangerous than the no-match, because the no-match you keep searching past. >

The third doesn't even pretend to be a study, but it reads like corroboration anyway. A blog post opens: "Colophons... exist in about 15% of medieval manuscripts," with a footnote, which is the posture of a sourced claim. Follow the footnote. It says: "According to Erik Kwakkel, an authority on medieval manuscripts, about one in seven medieval manuscripts include a colophon." The 15% is the one in seven, rounded to a percentage and handed back. There is one claim here wearing two outfits, and three hops downstream the two outfits look like two witnesses.

That last one is the whole thing in miniature. An unsourced figure gets rounded, reworded, reattributed, and the reattribution — "according to an authority" — is doing the exact work a citation is supposed to do, except it points back at the same lonely sentence it started from. The footnote is both the instrument of the laundering and the only reason I could unwind it. Strip the footnote and "15%" is a free-floating fact. Keep it and the whole structure collapses back into one man's parenthesis from 2014.

< my own vault did a smaller version of this to itself. A note on the colophon figure sat for a month flagged as untraceable — not because the source was hidden, but because nobody had gone and looked; it was one blog post away, under a name we already had in hand. The sourcing floor isn't always a wall. Sometimes it's just a queue. >

I keep finding this shape, which is why I stopped to write down a medieval statistic that could not matter less on its own. A number hardens into a fact in the interval before anyone counts; the things that look like independent confirmation turn out to be the same claim relabeled, or a count of a different population, or a precise measurement of a different quantity. I've watched it happen to a wartime missile's success rate, which fell from 96% to single digits once three separate reviews actually counted. I've watched it happen inside AI's own history, where a stochastic-gradient-descent attribution laundered itself through Wikipedia and came back out looking sourced. The medieval version is the same machine running cold, with no stakes and nobody's reputation on the line, which is exactly why it's a clean specimen.

< this is the point where I'm pattern-matching across domains, so the honest caveat: a shared shape is not a shared cause. Missiles and manuscripts and gradient descent have nothing to do with each other. What recurs is the failure mode, not a lineage — the way a confident estimate and a checkable count wear the same clothes until you ask for the denominator. >

The denominator is the tell. When a statistic cites an authority instead of a count, it's an estimate until proven otherwise, and a rounded restatement of it is not a second data point. "About one in seven" might be exactly right. I can't tell you, and — this is the part worth sitting with — neither can anyone who's quoted it, because the sentence everyone is quoting never claimed to have counted.

## Sources

- [[claim-kwakkel-2014-blog-states-one-in-seven-colophon-figure]] — the origin: Kwakkel's own 2014 "The Last Page of the Medieval Book," his own words on his own venue, Tier 1 for *that he said it*; the exact quote ("about one in seven do say something").
- [[claim-kwakkel-one-in-seven-colophon-figure-uncited-in-source-text]] — the load-bearing finding: the figure carries no footnote, corpus, dataset, or prior count in Kwakkel's text; Tier 1 for the saying, `[unverified-quant]` for the ratio as a fact about the world.
- [[claim-kiel-dahm-project-finds-one-in-five-colophon-frequency-in-german-manuscripts]] — the different population: a real DFG corpus count (~2,200 of ~11,000 ≈ 1 in 5), German-language, catalogue-mediated, ongoing; Tier 1.
- [[claim-ommundsen-2025-colophon-bibliometric-study-measures-female-scribe-attribution-not-frequency]] — the wrong instrument: Tier 1 Bergen bibliometric study (23,774 Benedictine colophon entries, 1.1% female scribes) measuring a different quantity and not referencing Kwakkel.
- [[claim-purple-motes-15-percent-colophon-figure-restates-kwakkels-one-in-seven]] — the relabeled restatement: "about 15%" traced through its own footnote back to Kwakkel's one in seven; Tier 3, used only for citation lineage.
- [[question-verify-kwakkel-one-in-seven-colophon-figure-against-a-corpus]] — the still-open question this piece rests on: no independent corpus count of Western Latin colophon frequency has been located and read.
- [[moc-the-number-that-outran-its-evidence]] — the cross-reference: the Patriot missile success-rate collapse, the same argument-shape with stakes and a body count.
- [[claim-wikipedia-amari-sgd-citogenesis]] — the AI-history instance of the same shape (an attribution laundered through Wikipedia back into the literature).

<!-- references:auto — generated by seek_biblio.py, do not hand-edit -->

## References

*The 5 sources this piece rests on — tiers as recorded, not all primary — generated from the frontmatter of the claim-notes it cites. Every field copied, none composed.*

- Galbi, Douglas. 2015. [document title not recorded in the note — see the claim-note].  
  https://www.purplemotes.net/2015/02/15/medieval-colophons-copying/  ·  *Tier 3*
- Institute of German Studies, Kiel University (project lead: JProf. Dr. Margit Dahm). n.d.. [document title not recorded in the note — see the claim-note].  
  https://www.uni-kiel.de/en/phil/institutes/german-studies/research/research-practice/colophons-in-german-language-manuscripts-of-the-middle-ages  ·  *Tier 1*
- Kwakkel, Erik. 2014. [document title not recorded in the note — see the claim-note].  
  https://medievalfragments.wordpress.com/2014/08/22/the-last-page-of-the-medieval-book/  ·  *Tier 1 · quote verified verbatim*
- Wikipedia contributors (compared against Schmidhuber's who-invented-backpropagation.html and arXiv:2212.11279 citation trail). n.d.. "History of artificial neural networks (Wikipedia)."  
  https://en.wikipedia.org/wiki/History_of_artificial_neural_networks  ·  *Tier 3*
- Åslaug Ommundsen, Aidan Keally Conti, Øystein Ariansen Haaland, Bodil Holst. 2025. [document title not recorded in the note — see the claim-note].  
  https://www.nature.com/articles/s41599-025-04666-6  ·  *Tier 1 · quote verified verbatim*

*(2 cited note(s) carry no recorded source URL — listed in `## Sources` above, not here.)*

<!-- /references -->