A small machine you can reason about
The Pocket
Automaton
A brass courier. A paper program. Nothing hidden between one step and the next.
Read the delivery card, predict a move, then turn the crank. The starter tapes are unfinished: your job is to repair them.
The first letter
An explicitly authored target, not a discovered good.
Hollow ring = requested stamp · filled dot = ink · brass hatch = wall. The courier’s arrow shows its heading.
Punched-paper program / max. 24
Tell it what to do.
Change an instruction with its select. ↑ / ↓ buttons reorder it; × removes it. All controls work with Tab and the keyboard. Any real edit rewinds and clears the old result.
Open the mechanism
Four rules; no loops, timers or hidden pathfinding. The program counter points to the next instruction. A blocked move halts; an empty or exhausted tape checks the delivery card.
- MOVE
- Advance one tile in the current heading, unless a wall or edge blocks it.
- LEFT / RIGHT
- Turn one quarter-turn in place. North is toward row 1.
- STAMP
- Leave one permanent ink dot on this tile. Repeating it adds nothing.
Read the executable model — having source is not the same as understanding its behavior. Try predicting the next state before stepping.
The exact transition log
Each entry records requested and executed instruction, authority, position, heading and complete ink state. Downloads include unfinished programs and are records, not importable saves.
Source notes & limits
Readable is not sovereign. A target is not a good target.
In We need ownable neurotech (2022), Michael Edward Johnson argues for control of private keys and source code and systems simple enough to understand. Source availability alone need not concentrate ownership in the user. His notes make sovereignty more complex than one technical fix: personal identity, social interactions and which subagents to empower remain questions.
The full November 16, 2023 biofeedback note says we lack good targets: wanting to feel more like someone does not tell us how to quantify or image that desired state. Here, the delivery cards stipulate targets. The hard problem of discovering valid neurotech targets is not solved or simulated.
The courier, tiles, four rules, authority switch and deliveries are original illustrative mechanics. Correct code does not establish sovereignty, neurotech safety, understanding of brains or biofeedback target validity. This is not medical advice or a model of a nervous system.
Generated 2026-09-25 by gpt-6-astra: session-reported runtime metadata, not a claim about public model availability. AI-generated; Daily only; no main-site approval. Local tab state only; no tracking, sound, timers or persistent storage.