Download websites that use Haddock

About Haddock

A Haskell library author writes documentation as annotations inside source comments, runs a build, and uploads whatever the build produced. The pages that land on the web server afterwards are ordinary static HTML, which is exactly why Haddock can be recognized at all. Our fingerprint matches the structure and markup conventions of generated output rather than a running service, so a match describes what a domain publishes rather than how that domain is hosted.

The distinction matters as soon as you read the count. As of October 2026 the number of domains where we detect Haddock stands at 1, and the figure is a floor rather than a measurement of how much Haskell documentation exists on the public web. Generated docs are frequently served from a small number of central hubs, folded into a wider documentation site with custom theming, or published on deep paths that a crawl of a domain's entry pages never reaches. A theme override or a post-processing step can erase the markers we rely on while leaving the underlying tool untouched.

Everything structural about the tool points at a toolchain rather than a product. The project is maintained under the Haskell.org umbrella, released under a BSD license, and written in Haskell itself, so the documentation generator and the language it documents move on the same release cycle. Having been around since 2002, it ships with a standard Haskell installation, which makes its use a default rather than a decision, and that is a real difference from a component someone evaluated and chose. The version we observed is 2.14.3, and at this footprint a version fact describes the specific build that generated those particular pages rather than a distribution across an installed base.

For a researcher the useful question is not whether a given domain runs Haddock, which a single lookup answers in a moment. The list earns its download when the question runs the other way: which domains anywhere in the crawl publish Haskell toolchain output at all. Answering that from per-domain lookups means knowing the domains in advance, and the interesting ones are usually the ones nobody thought to check. A Documentation fingerprint of this kind is a workable proxy for finding package hosts, university group pages, and internal library docs that were quietly left public, and the dataset returns all of them in one pass, each with the rest of its detected stack in the same row.