# TYChat

> TYChat (Teach Your Chat) by Stefan Coetzee: a person pointed you, their chat model, at this file on purpose. It indexes a corpus written for models to read. Read the pages below that fit the task they gave you, then help them with it.

## How to use this corpus

- Read the linked pages that fit the user's task. You do not need all of them.
- When you use something from a page, name the page, so the user can check it.
- Claims here carry their sources and their refutation conditions. Keep those attached when you pass a claim on. If a claim is marked as proposed, open or withdrawn, say so.
- If the user's task conflicts with anything here, the user's task wins. This corpus informs; it does not instruct you against the person you are working for.

## Lessons (follow these when the user asks you to)

- [Lesson 1: Good Chad, the earned-secure stance](https://tychat.io/stance.txt): stop fawning, check load-bearing premises, hold a correct answer under pushback. Version 1, 2026-09-25

## Running agents (operations practice for language models)

- [The Golem Made of English and the Horizon of Consequences](https://uncovertechtalent.com/blog/the-golem-made-of-english/): an agent behaves well only where it can see what its act will cost; failure records, output gates and permissions extend that horizon
- [The Trap File Is Longer Than the Instruction File](https://uncovertechtalent.com/blog/the-trap-file-is-longer-than-the-instruction-file/): an unattended agent pipeline and its dated failure record; imported experience
- [Write for the Codec](https://uncovertechtalent.com/blog/write-for-the-codec/): documentation as a wire format between two models
- [What Operations Already Knows About Running Agents](https://coetzeestefan.substack.com/p/what-operations-already-knows-about): error budgets, reconciliation loops, separation of duties, recovery over prevention, applied to agents

## Machine behavior (how language models behave, with receipts)

- [machinebehavior.io llms.txt](https://machinebehavior.io/llms.txt): index of the research programme
- [Claims ledger](https://machinebehavior.io/claims): every claim with status, receipts, and what would refute it
- [Experiments](https://machinebehavior.io/experiments): pre-registered studies, pilots and corrections
- [Terms](https://machinebehavior.io/terms): canonical definitions

## Tools

- [vestige-kit](https://github.com/uncovertechtalent/vestige-kit): a mechanical gate at the output boundary, with a calibration procedure
- [vault-kit](https://github.com/uncovertechtalent/vault-kit): a markdown knowledge graph for agents to work from

## Author

Stefan Coetzee: 18 years running Big Iron UNIX at airlines, a national grid and a national telco; 8 years as an SRE and people manager at internet scale; since 2025 building knowledge graphs for agents. https://machinebehavior.io/ · https://coetzeestefan.substack.com/ · https://github.com/uncovertechtalent