README grader

Paste a public GitHub repository. We read its README and tell you what it is missing, measured against 2,069 repositories we analysed for the README Benchmark. No signup, and the whole result is on the page.

What it checks

Six things, each rated Strong, Good, Needs work or Missing, with the evidence that produced the rating and one specific change to make.

Saying what it is. How far a reader travels before the first line of actual explanation.
Who it is for. Whether anything identifies the intended user. Absent from 99% of the benchmark.
Getting started. Installation and a runnable example.
Showing the thing. A screenshot, diagram or recording. Badges do not count, and we classify them separately.
Depth for an evaluator. Documentation links, examples, how it works, feature list.
Signals it is alive. Licence, contributing guidance, somewhere to ask a question.

Why there is no score out of ten

Because we would have to invent it. “README score: 62” implies a measurement nobody took, and it tells you nothing you can act on.

What we can defend is a count against a fixed thirteen-item checklist, and where that count sits in a real distribution of 2,069 repositories. When this tool says a README is more complete than 62% of them, that number was computed the same way for both sides of the comparison.

No language model is involved at any point, which is also why this is free and why two runs over an unchanged README give the same answer.

Limits worth knowing

Detection is pattern-based and English-only. An unusually phrased section can be missed.
It reads the README, not the repository. A project with excellent docs elsewhere and a thin README will grade poorly, which is arguably the correct answer for a reader who only sees the front page.
Public repositories only. Private ones are invisible to the tool.
The benchmark is observational. A missing element is an absence, not a diagnosis. See the full limitations.