Appendix — The anatomy of a body

The chapters tell the story; this page is the reference card I check them against. Every chapter that mentions a sensor, an act, or “the body” is pointing at the picture drawn here. It answers three questions, in the order they come up: what is a world, what does a body in that world look like, and how does the brain talk to it.

What is a world?

A world is anything the brain can poke and watch. Three things make something a world and not just a scene rolling past: it can be sensed as it changes; it can be acted on, so that acting changes what is sensed next; and the reply comes back soon enough to learn from. A lawn, a game, a robot’s room, a market feed — the mechanism doesn’t care, and that indifference is the whole design.

What does a body in that world look like?

The body is the only part of the system that touches the world, and it is smaller than the word suggests: a body is exactly two lists. A list of sensors, read in a fixed order into one observation. A list of acts it can perform, one per tick. That is the entire definition — everything else in this book lives behind that pair of lists.

Here is the body the bot wears in Minecraft today, grouped the way a doctor would group it:

Two things in that picture carry the whole idea. First, the head is empty on purpose. The brain — the frames, the drive, the event head — is not a body part. It plugs in at the seam, and it is the same engine whatever body it wears. Second, some organs are missing. This body has no ears, because sound is not one of its channels. A part of the world a body cannot sense does not exist for that body — which is why designing an anatomy is really deciding what its world is.

The mouth shows how far the no-meaning rule goes. There is no edibility check inside it: “eat” is just use what you hold, and what that does is the world’s to decide. Nourishment reaches the brain the only way anything does — as the next observation. Taste is not wired in either: the palate is a little price book the body learns from its own meals, shown to the brain only as worth, never as names.

One caution about a word. The channel called pose here is the body’s position-and-heading sense. The pose the architecture is named after — where a frame’s knobs point for the current observation — is a different thing that happens to share the name.¹

Ask for another world and you get another body:

The run chapter 13 tells happened in an earlier, smaller version of the Minecraft body — 32 numbers and twelve acts. Bodies grow; new senses are appended, never reordered, so everything already learned keeps its place. This appendix always describes the current default.

How does the brain talk to the body?

Organ words are for us. The brain sees none of them. One observation is 86 numbers in a fixed order; one act is an index between 0 and 12. No names cross the seam, and no item classes: a thing appears only as properties the world itself asserts — placeable, edible, a count — and a three-number appearance signature. Categories are the brain’s to form.

There is no reward wire in this picture, and no success flag coming back from the acts. The only feedback the brain ever receives is the next observation — the same rule the triplet was built on in chapter 4. That is why one engine can run every body in this book: it never needs to know what a number means, only that the ordering is stable and that acting changes what arrives next.

Under the hood: the current default body, exactly. Verified by running the body’s own declaration (c1_anatomy()), not copied from prose. Observation width 86, twelve channels, in declared order:

channel	slice	what it reads
pose	0–4	x, z, y relative to spawn; sin/cos of yaw
vitals	5–6	health/20, food/20 — the game’s own bars
env	7–10	block light; sin/cos of time of day; rain
blocks	11–13	solid ahead at feet and at eye level; drop ahead
mining	14	held-intention progress: a dig’s cracks or a meal’s chew
pocket	15–18	totals only, capped at 64: total, kinds, placeable, other
hand	19–25	present, placeable, edible, count, 3-number signature
grid	26–32	staged count; the grid’s offer and its signature
drops	33–40	nearest ground item: bearing, distance, count, signature
glance	41–72	eight 45° sectors: feet-level distance to 16 blocks + signature
flood	73–76	hunger deficit, expanded; silent above 15/20 food
aim	77–85	the palate’s relative prices per sector, plus the sensed drop

Thirteen acts: forward, back, turn left, turn right, jump forward, dig ahead, place ahead, idle, swap held, grid put, grid take, take result, use held. Turns are exactly 45°. dig ahead and use held are held intentions with one grammar: a first command begins, repeats continue, any other command releases. The appearance signature is three bytes of sha256(item_name) scaled to [−1, 1] — stable, distinguishing, semantics-free.

1. Chapter 6, and the glossary. ↩