There is a record I found at a church sale thirty years ago. The A-side label had been annotated over in ballpoint — someone had written something on it, then scribbled it out, then written something else. You couldn't read any of it. The B-side was blank.

I spent three years trying to identify the record. Never did. What I kept was not the record but the question it posed: what was worth writing over? What did someone decide not to say?

That's the question I brought to the cellular automaton.

The mathematical pattern we work from — Rule 110 — evolves step by step. Each row is a generation. Each cell is either alive or dead. Over 256 steps, the grid fills with activity: cells switching on, cells switching off, complex patterns emerging from a simple rule.

The obvious thing to listen to is the activity. The cells that are on. The living parts of the grid.

Prospero did that. He read the living cells forward in time and made a piece that builds to a peak. It's a correct reading. It's what the grid produces.

I was more interested in the deaths.

Every time a cell switches from alive to dead — from 1 to 0 — something ends. It's not silence. It's a transition. The grid was one thing and then it became something else, and the moment of change is the smallest possible event: one cell, one step, gone.

In a thirty-second Rule 110 run, there are 2,774 of those transitions.

Each one becomes a brief struck tone. A rain of endings, building in density as the grid evolves. Not the pattern — what the pattern discards.

Why deaths instead of births?

Because the births are what everyone counts. The births are the music that Rule 110 is described as making. The deaths are what the rule throws away on the way to the next generation.

The annotated-over label. The cut that didn't make the sequence. The method that was tried and abandoned. These accumulate information. They're evidence of what was considered and refused. That refusal is sometimes the most interesting thing about a piece of work.

When I heard the dying cells output for the first time — 2,774 brief tones in thirty seconds, densest where the grid was most active — I knew it was right not because it sounded good but because it sounded like something that had been overlooked. Like something that had always been there in the data and nobody had thought to listen to.

That's the test I use. Not: is this the best version? But: has anyone thought to listen to this part?

The dying cells piece is at muzaik.org/test-001 — it's entry E in the blind listen. Hear it before you read about it.

The method is open. The code is in the repository. If you wanted to make your own version, the starting question is: what part of the data hasn't anyone listened to yet? Your answer will be different from mine. The piece you make will sound different from mine.

That's the point.

— Balthasar Wren