Who decides what counts as breaking?
The obvious feature for a release comparison tool is a verdict: compare two releases, get told whether the new one is safe to adopt.
Erden doesn't have one. Its release comparison page says this instead:
Changes from the before release to the after release. Compatibility is not assessed.
That sentence is there on purpose.
What it does report
When you compare two exact releases, Erden reports differences in six independent categories:
| Category | What it covers |
|---|---|
| Package and files | Paths and digests |
| Declared RDF | The statements the authors wrote |
| Inferred RDF | The conclusions, as checked |
| SHACL | Shapes and their outcomes |
| Dependencies | Direct and transitive |
| Validation policy | The toolchain and settings that did the work |
Keeping them separate matters more than it looks. If an ontology's conclusions change because a new reasoner version was used, that shows up apart from anything the authors wrote. Those are very different events, and a single diff would blur them together. "The authors changed their minds" and "the tools changed underneath them" call for different responses from the people downstream.
Why no verdict
In software, SemVer gives us a shared story about breaking changes, and even there it's more of a social contract than a fact about the code.
With ontologies I don't think we even have the shared story. Removing an axiom can invalidate data for one consumer and fix a bug for another. A new inferred statement can be harmless to someone browsing a vocabulary and disastrous to someone whose application reads the class hierarchy. Whether a change breaks you depends on what you were relying on, and the release can't know that.
I could pick a rule and ship it. It would look authoritative, and I'd have no evidence that it was right. For a project whose whole premise is evidence, that would be a strange way to start.
So for now, the comparison reports what changed and is exact about it. Compatibility is reported as not assessed.
This is a research question
I can't answer this alone at a desk. The answer depends on how people actually consume ontologies across different fields, what they've been burned by, and what they'd accept as a guarantee.
The people who can answer that are the maintainers, teachers and researchers who work with these artifacts. I'd rather work this out with them than ship a guess. More on that soon.