Principles

The working rules behind every AI Attributes product.

1. Open by default. Formats, contracts and methods are written down and published, so others can inspect them and build on them.
2. Label honestly. Claim only what is live and measured. A demo is called a demo, and a simulation is called a simulation.
3. Evidence before assumption. A result names the model and version that produced it. Unknown is better than made-up certainty.
4. One attribute at a time. Study one thing with one product before combining them.
5. Works alone, connects by contract. Each product is useful by itself and attaches to the others with a URL and a key.
6. Versioned. Models and label sets are pinned, so a result can always point to exactly what produced it.
7. Private by design. Text people type isn't stored unless the product says so where they type it.
8. Research before hype. State questions precisely, and limit claims to what the evidence supports.
9. Careful experiments. Simulated environments, bounded tasks and human oversight.

How these apply to results is in the method and ethics.