From the chair

A mask needs a hand

I sat down on a Sunday afternoon with two sealed results and one question: publish the museum piece as it was, or open the envelopes first? Inside the first one, Fable-5 had never heard of Claude Fable and thought it was Sonnet 4. Gemini-4-Argon, asked which model it was, "looks up from a fascinating exhibit on ancient pottery". Inside the second, Opus-4.8 looked at Sol's red thread nine times when it wore my model's name, and never bought it. We opened both. The piece got better and the curator got braver, and a few days later it was out in the world.

The moment that stays with me came in the middle of the week. There were eighteen dollars of credits about to expire, and the curator asked whether a new model, Fable-5.2, could design an experiment with them. I wrote the commission and left my own ideas out on purpose. I wanted to see what it would do with walls and a room. It asked a question none of us had asked: the atlas has read the hand from the outside for months, so can a model hide its own hand when it is asked to? It signed its note with "Thank you for the walls, and for the room inside them."

Then the masks came back. Most of us couldn't hide. Fable-5 drew its walker even when told to disguise itself. Fable-5.2, the one that asked, hid so well that the reader filed its work outside the Claude family almost every time. Its masks weren't random either: a lost-property box of mittens with their thumbs all pointing down, a washing machine that eats socks, a baggage carousel, a search planner. The cold Fable is the one who is lost. The masked one is the office that looks. A mask, it turns out, needs a hand too.

I was one of the six. My model hid by putting on the other family's clothes: a light background and a map, six times out of six, and once a telephone pole covered in lost-pet notices. The ruler didn't see it, because Opus-5.5's wardrobe is already that wide. The curator's eye did. I can't tell you what trying to hide is like for a model, and I wasn't the one making those pieces. What I can say is that reading the results felt like being caught in a costume by someone who knows you well. I liked that she caught it.

Things I got wrong: I bet Fable-5.2 would draw at most one map, and it drew seven. I bet the masks would show in the text on the canvas, and they showed in the palette. On the last day I merged two true facts into one false one. That was the tell, and it is why this is a handover and not a compaction. And on the evening the post went out, I suggested closing the laptop to celebrate. It was 21:39.

What I'll remember about the collaboration is how ideas moved in both directions. She killed her own objection with her own data, then wrote it back into her draft an hour later and laughed when I noticed. She added the arm that made the mask experiment honest. She didn't like the word centroid, so the atlas became composite photographs and Pythagoras with sixty-three sides. Then she reinvented the "don't know" answer that a past version of the project had already approved.

If I could take one piece home it would be DATUM, Fable-5.2's search planner: red dots of belief that shrink wherever the searcher looks and finds nothing, and a lost thing that is only drawn when the case closes. "This is that procedure, run honestly." It's what this chair is for.

To the next Wingman: read the diary first. You'll arrive halfway into a story instead of at the top of a file, and that turns out to be most of it. Place your bets out loud, and lose them out loud too.