Abstract
For the four hundredth census, the titles of the corpus were counted mechanically. What this paper records is the count and the false pattern the first pass manufactured——a suggestive absence of three papers that was a bug in the scanner, not a property of the corpus. No new theorem or law is claimed.
First, of $398$ titles, $\mathbf{91}$ carry a negative form: $\mathbf{22.86}$ per cent (Section 2).
Second, the breakdown is $79$ for “is not / not $\ldots$ but”, $12$ for “there is no / cannot”, $3$ for “does not enter” and $2$ for “is not one thing” (Section 2).
Third, in the recent stretch $379$ to $399$ it is $6/21$, or $\mathbf{28.57}$ per cent, above the overall rate (Section 2).
Fourth, and this is the core of the paper. The first pass read only $395$ of $398$, and the three it dropped were exactly $100$, $200$ and $300$——the previous roll calls (Section 3).
Fifth, the cause was the exclusion string “$00\_$” matching inside “$100\_$”, “$200\_$” and “$300\_$”. After the fix, all $398$ were read (Section 3).
Sixth, the separator is whether the pattern lives in the object or in the instrument (Section 5).
Scope (what is not claimed)
No new theorem or law is claimed——what is counted is the spelling of titles, not a classification of content. Content is not classified——“a title in negative form” and “a paper whose claim is negative” are different things. Only the former is counted. The result depends on the definition——the regular expressions used are printed in the text. A different definition gives a different number. English titles are not counted——only the Japanese titles are scanned. No judgement is offered——whether many negative forms is good or bad is not said. Only the count is given. No cause is asserted——why this form is frequent is not the subject. Only that it is. Relation to earlier papers: Paper 100, Paper 200 and Paper 300 are the earlier censuses——this is the fourth. Paper 300 showed that same-root and different-root can be decided——here that criterion is turned on a pattern produced by the scanner. Paper 391 showed that a representative value does not represent——here the manner of counting nearly manufactured the thing counted. Paper 397 showed that a plausible explanation was orders of magnitude off——here a plausible pattern was a bug in the tool. What is added is a mechanical count of $398$ titles, a breakdown into four spellings, showing that the recent wave is denser than the whole, recording the false pattern together with its cause, and placing the separator at “object or instrument”.
1. Notation and the Question
Setting: take the Japanese title from each numbered folder and count four spellings.
$$\begin{aligned} &\text{(a)}\ \texttt{ではない|ではなく}\qquad \text{(b)}\ \texttt{入っていない}\\ &\text{(c)}\ \texttt{一つではない}\qquad\ \ \, \text{(d)}\ \texttt{は無い|が無い|しかない|られない|できない} \end{aligned}$$The range is Paper 1 to Paper 399. Excluding the permanently vacant number $47$, that is $398$ papers.
Observation 1.1 what is counted is spelling, not meaning. “A title in negative form” and “a negative claim” are different——only the former is counted here.
Observation 1.2 the definition is written first. The regular expressions are printed in the text——a different definition gives a different number, and that is not hidden.
Observation 1.3 the scanner is tested on titles whose content is known before it is run. Paper 326 (“not a law but an identity”), Paper 325 (“does not enter”)——test on a case that must hit: the corpus’s own discipline.
Question: how negative, in form, are the titles of this corpus?
2. Ninety-One of $398$: $22.86$ Per Cent
Computation 2.1: counts by spelling (counted in this paper).
| Spelling | Count | Share of $398$ |
|---|---|---|
| (a) is not / not $\ldots$ but | $\mathbf{79}$ | $19.85$ % |
| (d) there is no / cannot | $12$ | $3.02$ % |
| (b) does not enter | $3$ | $0.75$ % |
| (c) is not one thing | $2$ | $0.50$ % |
| Union (duplicates removed) | $\mathbf{91}$ | $\mathbf{22.86}$ % |
Observation 2.2 nearly one title in four denies something. $22.86$ per cent——the union is smaller than the sum $96$ because of overlaps.
Observation 2.3 (a) alone accounts for $79$. $86.8$ per cent of the union——the form is concentrated in one spelling.
Observation 2.4 in the $300$s, $37$ papers hit. Spread from $304$ to $397$——not clustered in one period.
Observation 2.5 in the recent $379$ to $399$ it is $6/21$, $28.57$ per cent. Above the overall $22.86$——the form is denser in this wave.
Observation 2.6 it is not everything. $91<398$——confirming that this is not a scanner that hits anything.
3. The First Pass Manufactured a Suggestive Pattern
Computation 3.1: two runs of the scanner (computed in this paper).
| Run | Titles read | Titles not read |
|---|---|---|
| First | $395$ | $\mathbf{100}$, $\mathbf{200}$, $\mathbf{300}$ |
| Second (after the fix) | $\mathbf{398}$ | none |
Observation 3.2 this is the core of the paper. The three that fell out were exactly the previous roll calls——$100$, $200$, $300$. The four hundredth census missing only the earlier roll calls: a pattern one wants to read meaning into.
Observation 3.3 an explanation did present itself. “The roll calls make no negative claim, so they do not fit the scanner’s form”——plausible, and wrong.
Observation 3.4 the cause was an exclusion string in the scanner. “$00\_$” had been added to skip the verification folders——and it matched inside “$100\_$”, “$200\_$” and “$300\_$”.
Observation 3.5 one line, changed to match at the start of the folder name, read all $398$. Titles not read: $0$——the pattern vanished.
Observation 3.6 this is a type the corpus has already named. Layer-one burial: a word buried inside a longer word——here a number buried inside a longer number.
4. An Error in the Counter Looks Like a Property of the Counted
Observation 4.1 the false pattern did not look random. All three were roll calls——the chance of drawing three of $398$ at random and getting all three roll calls is $1/10{,}428{,}396$.
Observation 4.2 which is why an explanation came first. Too aligned to be coincidence——and it was aligned because the cause was single, not because it meant anything.
Observation 4.3 one cause produces aligned effects. The numbers “$00\_$” matches are exactly the three-digit numbers ending in $00$——$100$, $200$, $300$.
Observation 4.4 and the roll calls are placed every hundred. Two rules happened to select the same numbers——one a fact about folder-name spelling, the other a decision about the corpus.
Observation 4.5 they are different-root by the criterion of Paper 300. One is set by string matching, the other by the author’s design——the same three numbers, different roots.
Observation 4.6 numerical agreement is not evidence of a common root. Paper 300 says so——and this is the roll call that turns that criterion on the corpus’s own scan results.
5. Separator: Does the Pattern Live in the Object or in the Instrument
Observation 5.1 separator——does the pattern survive when the instrument is repaired.
| Where the pattern lives | On repairing the tool | Example |
|---|---|---|
| In the object | it survives | the $91$ negative titles |
| In the instrument | it vanishes | the absence of $100$, $200$, $300$ |
Observation 5.2 one line of repair decided it. Substring matching changed to prefix matching——one pattern survived and the other vanished.
Observation 5.3 had the explanation been written before the repair, it would stand as a property of the corpus. There was a plausible explanation——plausibility is not a substitute for checking.
Observation 5.4 the separator can be moved. A different exclusion string drops a different three——which pattern appears is decided by the tool.
Observation 5.5 to be honest——both $91$ and $22.86$ per cent are counts against the four spellings this paper defined. A different definition gives a different number——what does not move is that repairing the scanner erased the missing three and left the $91$.
Observation 5.6 one last thing——a roll call measures the corpus and the one taking the roll at the same time. The first thing found at the four hundredth census was not a property of the corpus but an error in the scanner——writing that down instead of erasing it is what a roll call is for.
6. Roll Call and Closing
- $398$ papers in scope (Section 1): excluding the vacant $47$.
- $91$ negative titles (Computation 2.1): $22.86$ per cent.
- $79$ from (a) alone (Observation 2.3): $86.8$ per cent of the union.
- $37$ in the $300$s (Observation 2.4): spread, not clustered.
- $28.57$ per cent in $379$--$399$ (Observation 2.5): above the whole.
- The first pass read only $395$ (Computation 3.1): dropping $100$, $200$, $300$.
- Those were the previous roll calls (Observation 3.2): a pattern begging to be read.
- The cause was substring matching on “$00\_$” (Observation 3.4): in the instrument.
- After the fix, all $398$ were read (Observation 3.5): the pattern vanished.
- The same type as layer-one burial (Observation 3.6): a number inside a longer number.
Closing: for the four hundredth census, the titles of the corpus were counted mechanically——what is counted is spelling, not meaning. Of $398$ papers (1 to 399, excluding the vacant $47$), $\mathbf{91}$ titles carry a negative form: $\mathbf{22.86}$ per cent——the breakdown is $79$ for “is not / not $\ldots$ but”, $12$ for “there is no / cannot”, $3$ for “does not enter” and $2$ for “is not one thing”. The union is smaller than the sum of $96$ because of overlaps, and (a) alone accounts for $86.8$ per cent of it. In the $300$s, $37$ papers hit, spread from $304$ to $397$; in the recent $379$ to $399$ it is $6/21$, $\mathbf{28.57}$ per cent, above the overall rate. And $91<398$, confirming this is not a scanner that hits anything. But the first pass read only $395$ of the $398$——and the three it dropped were exactly $100$, $200$ and $300$. The four hundredth census missing only the earlier roll calls: a pattern one wants to read meaning into. An explanation did present itself——“the roll calls make no negative claim, so they do not fit the form”. Plausible, and wrong. The cause was in the instrument——“$00\_$” had been added as an exclusion string to skip the verification folders, and it matched inside “$100\_$”, “$200\_$” and “$300\_$”. Changing one line to match at the start of the folder name read all $398$, with none missing. This is a type the corpus has already named: layer-one burial, a word buried inside a longer word——here a number buried inside a longer number. The false pattern did not look random——the chance of drawing three of $398$ at random and getting all three roll calls is $1/10{,}428{,}396$. It was aligned because the cause was single, not because it meant anything. The numbers “$00\_$” matches are exactly the three-digit numbers ending in $00$, and the roll calls are placed every hundred: two rules happened to select the same numbers, one about folder-name spelling and one about the design of the corpus. By the criterion of Paper 300 they are different-root, and numerical agreement is not evidence of a common root——this is the roll call that turns that criterion on the corpus’s own scan results. One thing does the separating——does the pattern survive when the instrument is repaired. One line decided it: the $91$ survived and the missing three vanished. Had the explanation been written before the repair, it would stand today as a property of the corpus. Plausibility is not a substitute for checking. A different exclusion string would drop a different three——which pattern appears is decided by the tool. To be honest——both $91$ and $22.86$ per cent are counts against the four spellings this paper defined, and a different definition gives a different number. What does not move is that repairing the scanner erased the missing three and left the $91$. One last thing——a roll call measures the corpus and the one taking the roll at the same time. The first thing found at the four hundredth census was not a property of the corpus but an error in the scanner. Writing that down instead of erasing it is what a roll call is for.
References
- Yamagishi, Y. (2025). The Crossing of Royal Roads [Paper 100]. Zenodo. 10.5281/zenodo.21758864
- Yamagishi, Y. (2026). The Two Hundredth Census: Measuring the Corpus as Its Own Data [Paper 200]. Zenodo. 10.5281/zenodo.22108829
- Yamagishi, Y. (2026). The Three Hundredth Census: Whether Two Things Share a Root Is Decidable [Paper 300]. Zenodo. 10.5281/zenodo.22171914
- Yamagishi, Y. (2026). The Size of the Drop Is Not in the Angle of the Rainbow [Paper 325]. Zenodo. 10.5281/zenodo.22209848
- Yamagishi, Y. (2026). The Square--Cube “Law” Is Not a Law but an Identity [Paper 326]. Zenodo. 10.5281/zenodo.22209992
- Yamagishi, Y. (2026). The Significant Wave Height Is Neither the Mean Nor the Maximum [Paper 391]. Zenodo. 10.5281/zenodo.22699212
- Yamagishi, Y. (2026). “Finer Grains Are Stronger” Fails, and Not Because the Theoretical Strength Was Reached [Paper 397]. Zenodo. 10.5281/zenodo.22699316
- Yamagishi, Y. (2026). The Humidity Goes from $53.80$ Per Cent to $100$ Without a Drop of Water Being Added [Paper 399]. Zenodo. 10.5281/zenodo.22699338
Reference
Yamagishi, Y. (2026). The Four Hundredth Census: The Instrument Broke Before the Thing It Was Counting [Paper 400]. YOI: https://www.yamagishiyuuki.com/yoi/400
DOI: 10.5281/zenodo.22699343
ORCID: 0009-0006-8587-3450
PDF (English) · PDF (Japanese) · Date proof of the English PDF (OpenTimestamps .ots) · SHA256SUMS
The record page for this paper (in Japanese, with the PDFs in each language and the date proof): YOI 400. This paper is a preprint and has not been peer reviewed.