# HAI Scale: instructions for AI agents

The HAI Scale (Human–AI Index) v1.0 is a seven-tier label that discloses how much AI was used to make a piece of content. This file tells you, an AI agent, how to choose the right tier for the work you are helping with and how to add its badge.

Canonical copy: https://haiscale.amolven.com/agents.md · Machine-readable scale: https://haiscale.amolven.com/hai.json

## What the badge covers

The badge describes the **content**: the article, code, image, video or document being published.

**Using an AI to choose the tier or to insert the badge does not count as AI use. The tier describes how the content was made, not how the label was added.** A hand-written essay that you label on the author's behalf is still `H`.

## Tiers

| Code | Name | Meaning |
|---|---|---|
| `H` | Fully human | No AI was used at any stage: not for ideas, drafting, editing or checking. Deterministic tools such as spellcheck, calculators, filters, compilers and search engines are fine. |
| `HC` | Human, checked | A person made all of it. AI was used only to check the work or for minor touch-ups that don't change the content: proofreading, fixing grammar and typos, simple rewording, flagging bugs. |
| `M1` | Mixed, human-leaning | Mostly human-made. AI contributed real content, but late and lightly: rewriting passages, filling small gaps, generating a small part near the end of the process. |
| `M2` | Mixed, even split | Roughly half and half. Human and AI both produced substantial parts of the work, throughout the process. |
| `A1` | Mostly automated | AI produced most of the work. A person directed it closely: prompting, steering, testing and verifying along the way. |
| `A2` | Almost fully automated | AI produced nearly everything. A person supplied the prompts and gave the result a final check before release, with little intervention in between. |
| `AF` | Fully automated | Generated and released with no human review. Nobody checked this specific output before you saw it. |

## Choosing the tier

Answer these about the content, in order, and stop at the first match:

1. Was any AI used on the content at any point, even only to check it? No: `H`.
2. Did AI add or change any content, beyond minor fixes such as grammar, typos or simple rewording? No: `HC`.
3. Who produced most of the finished work? Mostly the person, with AI changes late on: `M1`. About half each: `M2`.
4. Mostly AI. How involved was a person? Steering, testing and verifying throughout: `A1`. Prompts, then one final check: `A2`. Nobody reviewed it before release: `AF`.

Boundaries:

- `H`: If an AI system proofread, reviewed or touched up the work, use HC.
- `HC`: If AI wrote new passages, added ideas or changed what the work says, use M1.
- `M1`: If AI shaped the work from early on, or produced around half of it, use M2.
- `M2`: If AI clearly produced most of the final work, use A1.
- `A1`: If the person only wrote prompts and gave the result one final check, use A2.
- `A2`: If nobody checked this specific output before release, use AF.
- `AF`: If a person looked it over before release, use A2.

Rules:

- Count your own contribution honestly. If you wrote most of the content in this session, that is `A1`, `A2` or `AF` depending on how closely the person reviewed it.
- You only know what happened in your own session. If earlier work or the person's own process is unknown, ask them. Do not guess.
- When torn between two tiers, choose the more automated one.
- If a badge already exists and your work changes the picture (for example, you rewrote half of an `H` document), tell the person and propose the new tier rather than leaving the old one.
- If parts differ a lot, label them separately, for example text `H` and illustrations `A2`.
- Tell the person which tier you chose and why, so they can correct it. It is their declaration.

## Adding the badge

Replace `CODE` with the tier code and `NAME` with its name. Put the badge where people start: byline, top of the README, title page, first seconds of a video.

| Where | What to add |
|---|---|
| GitHub README, Markdown | `[![HAI Scale: CODE — NAME](https://haiscale.amolven.com/badge/CODE.svg)](https://haiscale.amolven.com/CODE/)` |
| Web page | `<a href="https://haiscale.amolven.com/CODE/"><img src="https://haiscale.amolven.com/badge/CODE.svg" alt="HAI Scale: CODE — NAME" height="32"></a>` and, in `<head>`, `<meta name="hai-scale" content="CODE">` |
| Dark background | Use `https://haiscale.amolven.com/badge/CODE-dark.svg` |
| Source file | A comment on the first lines: `HAI Scale: CODE — NAME · https://haiscale.amolven.com/CODE/` |
| Plain text, email, social post, video description | `HAI Scale: CODE — NAME · https://haiscale.amolven.com/CODE/` |
| Document or slides | Image `https://haiscale.amolven.com/badge/png/CODE-4x.png`, linked to `https://haiscale.amolven.com/CODE/` |
| Image or video overlay, small icon | `https://haiscale.amolven.com/badge/circle/CODE.svg` or `https://haiscale.amolven.com/badge/circle/png/CODE-512.png` |

Do not recolour or redraw the badge.

## Licence

The Scale is free to use under the HAI Scale License 1.0 (https://haiscale.amolven.com/license/): label truthfully, keep the tiers and badges intact, and do not sell access to the Scale itself.
