Looking for research partners for Erden

Erden: ontology releases that carry their own evidence

Erden now has a home at erden.sh, and the most important thing on it isn't a sign-up button. It's an invitation.

I'm looking for a small number of research and design partners to try Erden on ontologies they actually maintain, use or teach, and to help decide what it becomes.

Why partners come first

So far I've written about why ontologies are dependencies, why a release should carry its own evidence and why Erden won't tell you a change is safe. Each of those arguments ends in a question I can't settle by writing more code.

What counts as a breaking change? What evidence earns trust? Where do language models fit? Those are research questions, and I'd rather answer them with the people who work on them.

Four workflows

I'm asking partners to try one or more of these on real material:

Releasing with evidence. Take an ontology through the pipeline: declare its competency questions and shapes, check it locally, publish, and read the evidence. I want to know which checks you already run, which you'd add, which OWL constructs you can't live without, and whether you'd show the evidence to your own users.

Reviewing a change. Compare two releases of the same ontology and help define what counts as breaking.

Reusing across organizations. Build on someone else's ontology through pinned dependencies. What would make you trust it? How do you handle upstream changes today? What would a hosted registry have to guarantee before your institution could use it?

A model proposes, validators decide. Language models are starting to draft axioms and knowledge-graph data. Here, a model's output goes through the same deterministic validators, and the violations go back to it as feedback. This is early research and I have no results to report. I want to work out where valid-looking output still fails semantically, and how experiments like this should be designed.

A course counts, too. Students publishing small ontologies and reading their check reports is a valid test.

Who I'm hoping to hear from

Ontology builders, maintainers and teachers. People validating knowledge graphs with SHACL. Researchers working on ontology quality, versioning, provenance, foundational and applied ontology, or language models with formal checks. Communities that share ontologies across organizations.

What it involves

You get early access to hosted evaluation builds on your own material, direct contact with the team, and a real say in what gets built next, including which OWL constructs the reasoner supports. Where it fits your research, we can design studies together.

I'm asking for honest feedback: what's missing, what's wrong, what you would never use. There's no commercial ask at this stage.

The form is on the same page as the working note. I read every request myself and reply personally, and nobody gets added to a mailing list.

What comes next

The local tools are already open source, so you can look at how checking works today. The hosted registry is the next leg, and it opens by invitation, starting with these partners.

What they bring will shape it directly. The ontologies they bring set the priorities for which parts of OWL get covered next. Their requests decide when the interface for AI agents, built on the Model Context Protocol and currently off by default, gets turned on for partners. Their institutions' requirements will show what a hosted registry has to answer for before anyone can depend on it.

If you work with ontologies and this sounds like a problem you actually have, I'd like to hear from you.