Evaluating documentation

Understanding documentation evaluations and how to request them

Documentation in the registry show evaluation scores that tell you whether they actually improve agent code quality.

What evaluations measure

Evaluations test agents on real tasks with and without documentation:

  • Does the agent use correct APIs?

  • Does it follow best practices?

  • Is the code maintainable?

Example: Without docs, an agent might use http.createServer() directly. With good docs, it uses Express's app.listen() - the intended API.

Viewing evaluation results

In the Tessl Registryarrow-up-right, evaluation results show:

  • How much a tile improves agent accuracy

  • Which tasks it helps with

  • Whether it's worth installing for your use case

Use these scores to choose tiles that will actually help your team.

What scores mean:

  • 3x+ improvement: Excellent for newer or niche packages

  • 1.5x improvement: Good improvement for most packages

  • 1x improvement: Minimal help (library already well-known)

Requesting evaluations

Need documentation for a public package that doesn't have coverage yet?

  1. Request an evaluation for that package

  2. The community gets access to evaluated documentation

This helps everyone using that package get better agent support.

Creating evaluated tiles

When you create and publish documentation, they can be evaluated:

Published tiles are evaluated to show their impact on agent accuracy.

What makes effective documentation

High-scoring tiles typically:

Show the right APIs

  • Public interfaces, not internal implementations

  • When to use which API

  • Common patterns

Cover real use cases

  • Tasks developers actually do

  • Complete examples, not just signatures

  • Error handling and configuration

Stay current

  • Latest library version

  • Breaking changes noted

  • Updates when APIs change

Next steps

Last updated