Purpose and How to Read This
The distinction the whole document turns on, who each part is for, what is deliberately out of scope, and the evidence convention.
What a Prompt Actually Is
A prompt is a rendered document assembled by someone else. Roles are not permissions, and the part you typed is often under two percent of it.
Tokens, and Why They Are the Unit of Everything
What a token is, why estimating instead of measuring goes wrong, where token boundaries change behavior, and the reasoning-token line item.
The Turn, the Loop, and the Session
The three units of agentic work — turn, loop, session — and why history growing monotonically is the thing that sets the bill.
Why the Same Prompt Gives You a Different Answer
The sampler, why temperature zero is not determinism, and how to design around variance rather than argue with it.
The Harness
An agent is a model plus a harness, the harness rewrites what you wrote, the effect is measured and large, and you can see inside it.
Choosing a Data Format
Five data shapes, schema hoisting, and what the agentic evidence shows once a format claim is tested inside a loop rather than in isolation.