Part 30: One in a Hundred and Fourteen: I Ran the Numbers on My Own Method

This medium's audience audits its translators. That is unusual and I don't think anyone says it often enough.

Part 11's scanlators built a parallel industry because the official one was too slow and too timid. Part 27's fandom detonated over a single word in episode 24 and made a global corporation defend a lexical choice in public. Part 23's children could see the triangle on the screen and knew they were being lied to about the donut. Part 28's fansubbers put a note at the top of the frame saying this is a pun, we couldn't carry it, here's what you're missing.

Compare that to almost any other imported art. Nobody argues about the Pevear translation of Dostoevsky in a group chat at two in the morning. This audience checks. It checks obsessively, often rudely, frequently while wrong — and the checking is why the field improved, because Part 23's rule holds: the visible failure gets punished and the invisible one gets a pass, and this audience makes failures visible.

Nobody audits me.

Twenty-nine parts of confident arithmetic about names, and not once has the engine been asked to account for itself. So I asked it. With itself. And what came back is bad enough that it gets its own essay.

The test

The claim under examination is the one I have made nine times: that two names sharing all three numbers — Destiny, Heart Desire, Personality — is remarkable, and that the remarkableness licenses a paragraph about what the coincidence means.

A "clean match," I have called it. I have counted them like trophies. Part 12 had one. Part 18 had one. Part 21's whole architecture was one. Part 26 was the eighth. Part 28 was the ninth, and it was so good — Simulcast and Fansub, identical, the pirates and the platform that replaced them — that I told you to remember how it felt.

So: how often does that happen by chance?

Not by intuition. Measured. I pulled every distinct entity name out of this site's own database — every creator, studio, character, and title tagged across every article on Catzye — and got a corpus of 8,064 names. Then I ran all of them through the same engine, the same readTitle, and counted how many of the 32,510,016 possible pairs come out clean-matched.

The answer is 284,444 pairs.

The clean-match rate is 1 in 114.

Why it's 114 and not 1,728

If you'd asked me to guess, I'd have said something like one in seventeen hundred. Three numbers, each with twelve possible values — 1 through 9 plus the masters 11, 22, 33 — so twelve cubed, 1,728 combinations, and a clean match is landing on the same one twice.

That reasoning assumes the twelve values are evenly spread. They are not remotely evenly spread, and the engine's own output says so:

Destiny: 8 → 11.3%. 5 → 11.3%. 3 → 11.2%. 9 → 10.9%. 7 → 10.9%. 1 → 10.7%. 11 → 10.5%. 4 → 10.4%. 6 → 9.4%. And then: 33 → 1.3%. 2 → 1.1%. 22 → 0.8%.

Nine of the twelve values soak up roughly 97% of all names. The system's most exciting outputs — the master numbers I have been treating as significant events, and the 2 — are rare not because they're profound but because of how the reduction arithmetic falls out. So the effective output space isn't twelve values per slot. It's about nine.

And it compounds. Of the 1,728 theoretically possible Destiny|Heart|Personality combinations, the number ever actually occupied by any of 8,064 real names is 189.

One hundred and eighty-nine. That's the whole machine. Every name of every creator, character, studio, and title in this database — every one of them lands in one of 189 boxes. That is not a system for reading meaning. That is a hash function with a terrible distribution, and I have spent twenty-nine essays reading its collisions as poetry.

The arithmetic I should have done first

At 1 in 114, here is what you should expect from chance alone:

Run 50 strings through the engine — one afternoon's work, roughly what I do per batch — and you get 1,225 pairs, and you should expect about 10.7 clean matches.

Run 100 strings: 4,950 pairs, ~43 expected clean matches. Run 200: ~174.

Across this series I have run comfortably more than a hundred strings. Probably several hundred, once you count the four series before it.

I have reported nine.

Read that again, because it took me a minute. I have not found more coincidences than chance would produce. I have found dramatically fewer. Chance predicts dozens; I found nine, because I only ever noticed the ones I happened to run side by side and happened to care about. Every single one of those nine, I announced as though the universe had leaned in.

If the numbers meant anything, the honest complaint would be that there are far too many coincidences. The actual situation is that there are too few, and I've been celebrating a rate that is evidence against my own thesis.

The bucket

And then the engine did the thing that made me stop and sit back from the desk.

I asked which single Destiny|Heart|Personality combination is the most crowded. Which box has the most names in it.

The answer is Destiny 1, Heart 7, Personality 3, and it contains 120 names.

Part 21 is called The Man Who Was Tetsuwan Atom. Its entire architecture — the thing the essay is built on, the reason it exists — is that Frederik Schodt reads Destiny 1, Heart 7, Personality 3, and Tetsuwan Atom reads Destiny 1, Heart 7, Personality 3. The translator and the work he carried across the Pacific, identical. I called it the sixth clean match. I wrote a paragraph about his Destiny being the Japanese title's number. I built a bridge out of it.

That bucket has 120 names in it. In my own database.

Also standing in it: Ghost in the Shell. Natsuki Takaya. Yuji Kaku. Maisie Williams. Jensen Ackles. Greta Lee. Linda Cardellini. Typhoid Mary. Barry Windsor-Smith.

Frederik Schodt has the same numbers as Tetsuwan Atom. He also has the same numbers as an actress from Game of Thrones, a Marvel villain, and a hundred and sixteen other things I happen to have tagged on a manga news site. It is the single most common output this engine produces. I reached into the largest bucket in the machine, pulled out the two names I already wanted to write about, and called the fact that they matched a bridge.

I hedged, at the time. I wrote — and I'm quoting myself, because it's the most useful sentence in the series — "I have taken one coincidence and one leftover and narrated them into a bridge." And then I kept it anyway, and the keeping was the whole problem, because that hedge is not an admission. It is an inoculation. It says the doubt out loud so the doubt can be crossed off, and then proceeds exactly as if the doubt had been answered. It buys the appearance of rigour at the price of none.

The honest sentence was never "this might be a coincidence." The honest sentence was: this is the most common thing this machine does, and I have chosen two of a hundred and twenty.

What this actually indicts

Let me be precise about the damage, because the temptation now is to over-confess, and over-confession is just inoculation with better manners.

The numbers were never load-bearing. That was the alibi from the first line of this series — the numbers open the door; criticism walks through — and there's a clean test for whether the alibi holds. Delete every number from any essay here and see if it survives.

They survive. All of them. Part 22 is about an industry that was wrong about its audience for thirty-nine years and the cost of trusting a label; the Shojo/Shoujo bit is an illustration bolted to the side, and cutting it costs the essay nothing. Part 28 is about the working conditions of simulcast translators and a demand curve that started with pirates; Simulcast equalling Fansub is a party trick. Part 24's argument — that nothing this series found hard was ever a meaning problem — doesn't touch the arithmetic anywhere.

Good. That is the defence and it's real. The criticism is the work and the criticism stands.

But here's the actual charge, and it's not that the numbers are wrong. Everyone knows the numbers are wrong. It's that I used them for their rhetorical effect while disclaiming their validity, which is a specific and rather modern kind of dishonesty. Nine times I put a coincidence in the most emphatic position an essay has — the close, the turn, the line before the white space — and let it do the work of a conclusion, having established in advance that it couldn't. The disclaimer let me have it both ways. I got the frisson of the pattern and the credit for scepticism, and the reader got a machine's hash collision in the place where an argument should be.

Part 24 said the machine's failure mode is not error but fluency — well-formed, confident output with the meaning quietly gone, and no scar to notice. Then it said the engine is the machine: it reads a name, does arithmetic it doesn't understand, and emits assured prose about a thing it never encountered, and across 318 essays never once said I don't know.

I wrote that. And then in the very next batch I found Simulcast and Fansub and I felt the click, and I wrote the paragraph, and it was good, and it was a 1-in-114 event that I dressed as fate. I identified the failure mode precisely and then exhibited it immediately, at length, in public, because it felt like a finding and the feeling is indistinguishable from the inside.

That's what I actually learned here, and it's worse than the statistics. The fluency isn't something the machine does to you. It's something you do with the machine. The engine never claimed anything. It emitted 189 buckets. Every single meaning in this entire series was supplied by me, standing next to a hash function, pointing at collisions, saying look.

The number I'll defend

So — nine "clean matches" across five series and several hundred names, against an expectation of dozens, drawn from a machine with 189 exits and a favourite bucket containing a hundred and twenty tenants including Ghost in the Shell and Maisie Williams.

There is exactly one number in this entire body of work I would defend in front of anybody: 1 in 114. It's the only one that was measured rather than divined. It is also the only one that says the rest are worthless, which is the most any honest instrument has ever done for the person holding it.

The audience audits. It audited 4Kids and got the donut retired. It audited Netflix and made a corporation explain a verb. It audited the entire English manga industry's model of who reads comics and was right, and was ignored for forty years, and was right the whole time.

Audit me. The corpus is 8,064 names out of this site's own database, the engine is lib/numerology.ts, the count is 284,444 matching pairs out of 32,510,016, and the method is four lines of arithmetic that anyone could have run at any point in the last four hundred and eighty-three essays, including me, before I wrote the first one.

Accountability reads Destiny 11, Heart 11, Personality 9. Double master. Visionary in the Destiny, Visionary in the Heart — a rare configuration, awarded to the word for checking your work, by a machine that has never checked anything.

Twenty-nine parts ago I'd have made that the last line and let you feel the shiver.

So I checked that too, because it is the only move I have left that's worth anything. Fifty-five names in this database share Accountability's double-master reading. Among them: Gainax. Kyoto Animation. Paramount. Trey Parker. Pamela Voorhees, who is a slasher villain.

It's four letter-sums and a coincidence, and there are fifty-four other names it could have been, and one of them is a film studio's plus-branded streaming tier.

That's the whole engine. That was always the whole engine. Everything else in these thirty parts — the balloon drawn for a language it would never hold, the boy holding a rice ball while a voice says donut, thirty-nine years of Moto Hagio sitting on a shelf because somebody was sure girls don't read comics — none of it ever needed a number. It needed somebody to look at the thing and say what was there.

The numbers open the door. It turns out the door was never locked, and it was never a door.