The Fish Trap

Four years and 300 pages later, I have to decide whether to let my brother read it to you.

Nothing in this one will help you migrate anything. No Graph endpoints, no pipelines, nobody’s LMS gets a kicking. Normal service resumes soon. I promise.

But it is about the thing this publication is named after, so stay if you can stand it. My brother Daniel died in September 2021. I’ve spent four of the five years since writing a book about him. It’s around 300 pages and it’s almost finished. Before I even consider an audio book version I have to make a decision about AI voice models, and I’ve ended up working that decision out with the help of a Chinese philosopher who has been dead for roughly 2,300 years. That’s the post.

A plastic table with holes drilled through it

In late 2018 I converted a room under my house into a recording studio. Calling it a studio is generous. The previous owners had roughly constructed a subfloor space that was supposed to be a bedroom and never quite got there, and I filled it with a foldable plastic table my parents had bought me, four cheap office chairs, a mixer, and a PC I had lying around. I drilled holes through the table so the mic cables could run underneath, then threw a tablecloth over the top so nobody on camera would work out what was under there. The chairs broke almost immediately and stayed in service for two years anyway.

The microphones were the best I could afford at the time, because audio matters and some part of me already knew it was going to matter more later. I still have them.

Daniel called the room “Mum’s basement” on air. It was my house. I never bothered correcting him.

The first episode of Invert the Why went out on 20 January 2019. Daniel named it, after the control setting he used in every game he ever played, where you push forward to look up and pull back to look down. Most people never touch that setting. He swore by it. We recorded 69 full episodes over two and a half years (and one 20 minute intro that I still count separately, don’t ask), the first thirty-odd of them in that room.

What the files turned out to be

In 2019, a recording like ours was a fixed object. Four idiots, four microphones, one small room with hard walls, everybody bleeding into everybody else’s channel. If you wanted one voice on its own you were out of luck, and you certainly weren’t getting it out from under an office chair that creaked every time its occupant laughed.

That has changed, and mostly in the last 18 months or so. Software will now work out who spoke when, strip the room and the reverb and the chair off a track, and pull a single voice out of a mix where people are talking over the top of each other (Adobe’s Enhance Speech and iZotope RX both do good work here, and the ElevenLabs isolator is the one I keep coming back to). That last part still falls over on our worst overlaps… but it gets there more often than it doesn’t. Direction of travel rather than finished product, and the direction isn’t in doubt.

None of this generates anything. It recovers what was always on the tape and was previously sitting underneath three other idiots talking over the top of it.

Which means I can make an audiobook where I read my chapters and Daniel reads his.

The book already half does this on paper. Every one of his stories is tagged with the episode it came from, and there’s a QR code at the back pointing at the feed, because I wanted readers to be able to go and hear him tell it properly. His words sit indented and italicised so you can always see the exact point where I stop and he starts. In audio that indent stops being a typographic convention and becomes a cut to tape. His timing. His soundboard buttons. His inability to finish a sentence without three asides, a callback, and a loose sprinkling of profanity.

The thing I keep not doing

The other option is the one people assume I’d take. Clone the voice, feed it the manuscript, have Daniel narrate the entire book including the 51,000 words he never said. The tools are good enough now that it would take an afternoon and most listeners would never catch it. (ElevenLabs is still about the best for this, and it isn’t close imho.)

I’m not going to do it. I’ve had one rule since the first draft, which is that Daniel’s words are untouched. Not tidied in any way, not smoothed, and not corrected for grammar or for the fact that he swore in roughly every second sentence. A cloned voice would let me keep that rule on paper and destroy it in practice, because a listener would have no way of telling which sentences came off the tape and which came out of me. The archive is the only part of this project I can’t fake. Faking it is the one move that would cost me the whole thing.

I spend a lot of my working life arguing that institutions should own their data rather than rent it as a feature of somebody else’s product. Turns out I have a personal version of that argument, and it’s 300 pages long.

Zhuangzi drums on a tub

Somewhere in the second year of writing I started returning to Chinese philosophy (which I read a lot of when I was younger), mostly because the Western grief literature had stopped being useful. It kept arriving in stages, and my grief has never once done anything in a predictable sequence.

Zhuangzi’s wife dies in chapter 18 of the book that carries his name (older editions have him as Chuang Tzu, if you ever go looking for a secondhand copy). The chapter is called Perfect Happiness, and it also contains the passage where he interviews a skull, who tells him that being dead is better and declines the offer to be brought back. So it’s a strong chapter. His friend Huizi turns up to pay respects and finds him sitting on the floor with his legs sprawled out, banging on a tub and singing. Huizi is appalled and says so, at length. The bit most retellings leave out is Zhuangzi’s answer, which opens with him admitting he howled like anybody else when it happened. The singing came second. It came from working something through and out the other side, rather than from feeling less. Nobody in that story is composed because they loved less, which is more than I can say for most of what I read in 2022.

Then there’s chapter 26, External Things, which now reads as a joke at my own expense. The bit I mean sits right at the very end of it. The fish trap exists because of the fish, and once you’ve got the fish you forget the trap. The snare exists because of the rabbit, and once you’ve got the rabbit you forget the snare. Words exist because of meaning, and once you’ve got the meaning you forget the words. Zhuangzi finishes by asking where he might find a man who has forgotten words, so that he can have a word with him.

Something I only worked out properly last year, and it has been irritating me ever since. The book comes in three parts. The Inner Chapters, one to seven, are the ones scholars generally accept as the work of Zhuang Zhou himself. The Outer Chapters run eight to 22 and the Miscellaneous Chapters run 23 to 33, and both are largely later followers writing in his voice. Which puts the tub in the Outer and the fish trap in the Miscellaneous. Neither of the two best bits are reliably his.

I have spent four years being fanatical about which sentences in my book are Daniel’s and which are mine. Indenting. Italicising. Refusing to tidy a single word of his. And the most useful thing anyone has said to me on the subject was probably written by somebody else, wearing a dead man’s name, about 2,200 years ago. I don’t know what to do with that, so I’m leaving it here. (Brook Ziporyn’s 2020 translation is the one I’d put in your hands, mostly because his notes are honest about all of this. Burton Watson if you want the older standard.)

I named this Substack after the idea that good scaffolding exists to be discarded. Build the capability, then get out of the way. “Teach yourself out”.

Then I spent four years building the largest fish trap of my life. It runs to 300 pages, with 69 episodes of supplementary fish trap attached. Whether I caught the fish is the one thing the book can’t tell me.

Mozi thinks I’ve wasted four years

In Analects 17.21 a student called Zai Wo puts it to Confucius that three years of mourning is excessive and one year would do. Confucius asks whether he’d feel comfortable eating good rice and wearing fine clothes after a year. Zai Wo says he would. Confucius tells him to go right ahead, then waits until he’s out of the room before saying what he actually thinks of him.

What strikes me about the three-year period described in Chinese history is that it was structured, public, and required of you. You wore particular clothes. You stepped back from official work. Everyone around you could see the state you were in, because displaying it was the obligation. In the modern west (and as it’s usually portrayed in film) you get a fortnight of casseroles followed by a broad social expectation that everything should be back to normal. I’ve come round to thinking the casseroles are a shitter system.

Mozi disagreed with all of it. His essay on moderation in funerals argues that elaborate mourning impoverishes families, weakens states, and is performed at least partly for an audience. He’d have looked at four years and 300 pages and told me to get back to work. I don’t have a simple answer for him, and on some mornings I think he’s right.

Mostly though, I think he can get f*cked.

Voice and countenance

There’s a phrase you’ll find on Chinese funeral wreaths and memorial scrolls. 音容宛在, yin rong wan zai. The voice and the face seem still present.

It reads as consolation, and it’s also an admission, because it was written by people who knew perfectly well the voice was gone and were describing the sensation of almost being able to hear it. The spirit tablet works on the same principle. You write the name on a piece of wood, put it on the altar, speak to it, tell it the family news. The dead get relocated into an object and a set of obligations you keep up for the rest of your life.

I have 69 episodes. Hundreds of hours. I don’t need to write his name on anything, because he’s on Spotify, mid-sentence, arguing with Rocket Russell about Formula One and losing.

Two and a half thousand years of people wishing they could still hear the voice, and I’ve got mine as a file. I’m still working out whether those are the same thing.

What four years of this actually taught me

The book was never for me. I already have Daniel, in the way anyone has anyone once they’re gone, which is partially and unreliably and mostly at three in the morning. The book is for people who never met him, and it works on them, and watching it work has been the strangest part of the entire exercise. Strangers have opinions about my brother now. They quote him. One of my early draft readers told me she’d started the podcast on her commute and had to stop, because she was laughing too loudly on a train.

The audio is the part that’s for me, which is probably why I keep circling around the audiobook without starting it. Publishing the book was fine. Publishing his voice feels like a different thing, because a book can be shut while a voice arrives in your ears whether you were ready for it or not.

Zhuangzi says get the fish and forget the trap. The book comes out on 10 September 2026, five years to the day. After that I find out whether I can put the trap down.