> For the complete documentation index, see [llms.txt](https://docs.wormbrain.info/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.wormbrain.info/explore-wormbrain/03-the-experiment/how-we-trained-it.md).
# how we trained it
The brain wiring stays the published wiring. No extra neurons. No synapses cut because they got in the way of a profitable habit. We changed how strongly a small number of cells listen, then watched whether the crawl through a market got less clumsy.
## What is allowed to move
Each cell has a gain: one number that says "when your neighbors talk, how much do you care?" Before training, those gains start even and dull. An outer loop is allowed to nudge them.
1. Run the worm through a stretch of tape (real prints from tokens it is allowed to see).
2. Read the command cells. Turn them into paper intents.
3. Let the cage accept or block those intents.
4. Write the result back as a shove the worm can feel, on cells that already handle a bump from behind (the PVC / DVA neighborhood).
5. Nudge the listening strengths a little. Keep the wiring table frozen.
Only those listening strengths are free. They are stored, named, and inspectable. If we publish a dump of the animal, the gains are in it.
## What is not allowed to move
* The list of who connects to whom
* The sign of a chemical contact once we have pinned it
* The names of the cells
* A hidden 303rd cell
* A hand-written rule that says "if RSI is low, buy"
If a change cannot be pointed to as a gain on a named cell, it is not training. It is a new experiment.
## What it is being trained toward
Not "maximum dollars." That target turns the project into an ordinary optimizer with a worm costume.
The outer loop scores a stretch on a short list:
* Did the command cells produce a clean forward, reverse, or pause, or did they chatter?
* When the cage filled an intent, did the next few observations shove the animal around less violently?
* Did the worm stop lunging at every wick?
Profit can go up when those things improve. It is not the only number we look at, and it is not a guarantee it stays up.
## Paper first
Training ran on paper fills first. The cage pretended a fee, a bit of slip, and a veto list. That kept a bad gain set from becoming a live hole in the wallet while we were still turning knobs.
{% hint style="info" %}
Think of it as teaching a worm that some smells come with a poke, not as teaching it finance. After enough pokes, ASH-side cells get a louder vote.
{% endhint %}
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:
```
GET https://docs.wormbrain.info/explore-wormbrain/03-the-experiment/how-we-trained-it.md?ask=&goal=
```
`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
