COMMAND

How do I make the machine useful without trusting it blindly?

After this station you can turn a vague AI request into a controlled command — a reusable spec that makes the machine work for your investigation instead of inventing it.

// SIGNAL

The machine answers anything, instantly, fluently. That is exactly the trap — it will finish your sentence before you have finished thinking.

Chatting gets you a confident wall of text. Commanding gets you usable work. The difference is not better vibes — it is better instructions.

// METHOD

A command is a contract: a role, the context, one task, hard constraints, what evidence it may use, the output format, and how to mark what it does not know.

1Role — tell the AI what to act as: an investigator, not a hype-man.

2Context + Task — what it needs to know, and the single thing it should do.

3Constraints — what it must not do; force it to separate facts, assumptions and unknowns.

4Output + Verification — the exact format, and where to flag uncertainty or missing evidence.

// TOOLING

Any capable AI assistant works. You command it; you do not believe it.

Any AI assistantYour G1 questionA plain doc for the log

// MISSION

Take your question from G1 and turn it into three controlled commands — then prove you improved them.

1Write an Exploration command: map possible actors, data sources and angles — facts vs assumptions vs unknowns, explicitly.

2Write a Structuring command: turn messy notes into a table or outline.

3Write a Verification command: flag weak claims and missing evidence.

4Run them, then log what changed between your first prompt and the final usable one.

// VERIFICATION

A command is good enough to keep when:

→It separates facts, assumptions and unknowns — not one confident blur.

→It names what the AI may and may not use as evidence.

→Its output is structured enough to reuse next time.

// GHOST

The machine wants to complete the pattern. It does not know when the pattern becomes a lie.
Your job is not to admire fluency. Your job is to command the work.
The prompt is not a spell. It is a contract.

// BRIDGE

A command is only as good as what you feed it. Next — INTAKE: the raw material worth commanding the machine with.

YOUR ARTIFACT

Command Log

Your three commands, the problems you caught, and the final reusable prompt — proof you can direct the machine, not just ask it.

SELF-CHECK 0 / 5

Begin the mission →

ARTIFACT TEMPLATE

# Command Log — [Investigation Title]

## Starting Question
The question from G1.

---

## Command 1
- **Goal:** What did I need the AI to help with?
- **Input:** What context did I provide?
- **Command Version:** v1
- **Constraints:** What did I forbid or require?
- **Output:** What did the AI produce?
- **Failure:** What was wrong, vague, invented or incomplete?
- **Revision:** How did I improve the command?
- **Useful Result:** What survived?
- **Verification Need:** What still must be checked?

*(Repeat the block for each version: Command 2, Command 3 …)*

## Reusable Command
The final prompt template, ready to reuse.

Prompts are not vibes. Prompts are working specs.

NEXT STOP ON LINE G

G3 INTAKE OPEN · LIVE NOW