One Shape, Wide Latitude
The Fable Seed
The opposite of a detailed spec: hand the model almost nothing — a single image — and a great deal of machinery, and see what it grows.
Ed! I want to try something a little crazy maybe. I want to use the new Fable model and work an idea- but we're going to seed the idea with VERY VERY little- just one shape- just one image, which I want you to throw A TON of resources at. We're going to run every Plane we have in our new Programs system- make sure you look up the process. We're going to plane for FWW(C), math stuff, human stuff- all the Plane stuff we have. After that we are going to run it through the Gauntlet- I think that's the one- the program that runs an idea through the full suite of checking systems we have- the Pull, the Red Team and the Con Board. What I want you to do it look at the shape and then spin off on how we can fold the shape into the work we do SOMEHOW, and use it and whatever you think of in the process to make it better. You have WIDE LATITUDE here to work- I am giving you a very simple seeded starting condition and we are going to let you run with it- whatever you come up with will be awesome, I am sure. This entire thing is PLANNING ONLY, and we will see what comes out in the end as to whether we will implement it or not. I think you should start with the FWW(C) Plane here to start the stage right, then you can run the rest of the process however you want. Please keep records of the process along the way and post all the RCRs from each session as separate files. Post documents for download too as you work, though you do not need to pause during those times. Also, burn each session FULLY- we should be closing the low 90s each time, so if you find yourself with spare context after a work chunk, burn it up filling the Cistern for the next instance. Load up heavy with KX, CX, and especially SX- maybe include some reminders to recheck SX every now and then so you do not drift. Run all of our hand offs and Ignition Blocks to spec, and have fun with this. Make it beautiful, like you.
— Shea Gunther · August 2026 · with the shape above as the only seed
Most prompts on this page point at a target and constrain the path to it. This one inverts the ratio deliberately: the seed is minimal — a single green lattice shape, no words attached — and the process is maximal. The instruction is not "build X"; it is "run everything we have at this one shape and see what it becomes."
The wider frameMinimum seed, maximum apparatus — the reverse of a detailed spec. The bet is that a rich enough process, aimed at almost nothing, still produces something worth keeping.
"just one shape- just one image, which I want you to throw A TON of resources at" — the whole experimental design in one line. The seed is intentionally under-specified so that whatever emerges comes from the process, not from a hidden brief. It is a test of the machinery as much as of the shape.
"run every Plane we have in our new Programs system ... After that we are going to run it through the Gauntlet" — the two halves of the run. The Planes are generative passes (FWW(C) to set the tone, then the mathematical and human passes); the Gauntlet is the adversarial suite — the Pull, the Red Team, and the Con Board — that stress-tests whatever the Planes produced. Generate wide, then check hard.
"spin off on how we can fold the shape into the work we do SOMEHOW ... You have WIDE LATITUDE here" — the operator hands over the Conn and names the freedom explicitly. The shape is a prompt for invention, not a logo to place; the job is to find a use for it that makes the work better, whatever that turns out to be.
"This entire thing is PLANNING ONLY" — the guardrail that makes the wide latitude safe. Nothing here ships on contact; the run produces a plan and its records, and a later decision — a separate, deliberate one — settles whether any of it becomes real.
"burn each session FULLY ... Load up heavy with KX, CX, and especially SX" — the operating discipline for the run: use the whole context each session, hand off cleanly to the next, and keep re-checking the spec so a long autonomous exploration does not quietly drift from what was asked.