feat(ontology): add Cohort Discovery rule deletion, fix stale URIs and SWRL raw-editor sync

[jeremiaspf]

What

Three unrelated fixes in the Ontology Designer's Cohort Discovery, Expressions & Axioms, and SWRL screens, grouped together because they're small and touch neighboring code.

Cohort Discovery had no way to delete a rule from the UI. The backend delete endpoint already existed; query-cohorts.js gets _syncDeleteBtn/deleteActiveRule and _ontology_cohorts.html gets the "Delete rule" button (next to "Save rule", disabled until a rule is loaded) to actually call it.

Cohort Discovery also had the same stale-URI problem described in the reasoning/generation PRs, but more visibly: nothing on this screen reconstructed a class/property URI from its current Base URI, so a re-based domain could make an apparently correctly-configured rule show 0 members/0 edges with no obvious cause. Adds base-URI tracking and a _currentUri() helper, used everywhere the screen previously read uri/iri/id directly (class selection, lookup-by-URI, available data properties, compatible relationship properties between two entities, each hop's dropdown).

Expressions & Axioms editor now prefers an element's simple name over its stored URI when populating class/property dropdowns (previously the reverse). The backend (OntologyGenerator._resolve_uri, see the companion generation PR) already rebuilds the correct URI from a simple name against the current Base URI — sending the name instead of a potentially-stale URI avoids axioms carrying a stale URI forward in the first place.

SWRL rule editor: the entity filter for the rule graph now also includes direct subclasses of entities that participate in a business relationship, even when the subclass itself has none — previously such a subclass couldn't be referenced or typed against in a rule, despite being a reasonable thing to do. The raw-text editor's auto-sync from the visual graph selection is now suppressed while the editor is in raw mode, so typing an atom with no corresponding graph node (e.g. a derived class in a THEN clause) is no longer silently overwritten. Also switches the D3.js CDN to jsdelivr (same change as the companion frontend PR — this file needed other edits too, so it's included here rather than split further).

Why

The delete-rule gap and the stale-URI issue both make Cohort Discovery awkward to actually use day to day. The axioms/SWRL fixes are UX papercuts that were easy to fix alongside the above while in the same area of the codebase.

How to test

1. Create a Cohort Discovery rule, save it, then use the new "Delete rule" button — it should remove the rule and the button should be disabled again with no rule loaded.
2. Rename a domain's Base URI after creating classes/relationships used in a Cohort rule — before this fix, the rule shows 0 members/edges; after, it resolves correctly.
3. Save an axiom in Expressions & Axioms referencing a class, rename the domain's Base URI, reopen the editor — the dropdown should still resolve the class correctly.
4. In the SWRL editor, reference a subclass with no relationships of its own in a rule — it should now appear in the graph. Switch to raw-text mode and type an atom with no matching graph node — it should no longer get overwritten by the visual sync.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[benoitcayladbx]

Review — not ready to merge

Looked at the four-file diff against develop. Direction is right (Cohort delete belongs in the designer; stale URIs on this screen are a real footgun), but a few issues should land before merge. I did not merge.

Blockers

1. _currentUri() always mints hash URIs — that disagrees with the rest of the stack (query-cohorts.js)

const base = (this.baseUri || '').replace(/[#/]+$/, '');
if (base && name) return `${base}#${name}`;

Ontology.ensure_uris and OntologyGenerator._resolve_uri do base_uri + localName (after ensuring the base already ends in # or /). CohortBuilder._to_ontology_uri picks the separator from the live base (# if the base ends with #, else /). Tests in tests/units/dtwin/ use slash bases (http://acme/ → http://acme/Person).

This helper would emit http://acme#Person for that ontology. Re-saving a Cohort rule after a rebase would then persist the wrong URI, which is worse than the stale-URI bug.

Please join the same way as _resolve_uri / ensure_uris: do not strip the terminator and force #. Fallback: if the base has neither terminator, append # then concatenate the local name.

Also remap already-saved class_uri / hop via / target_class by local name on load. _currentUri only helps new dropdown picks; a stored stale URI still will not match this._currentUri(c) === hop.via, so _renderHopRow will blank the hop instead of rewriting it.

2. _dataPropsForClass behaviour change looks like a regression (query-cohorts.js)

Previously: ontology-level data properties, filtered by domain, with a fallback to the full list.

Now: nested cls.dataProperties || cls.properties || cls.attributes, and [] if _classByUri misses.

Two problems:

• If the class lookup fails (stale URI on a loaded rule — see above), filters disappear instead of degrading.
• Nested dataProperties is only guaranteed after Ontology.sync_class_data_properties. If a payload still only has top-level properties[] with domain, the where-clause dropdown goes empty.

Keep the reconstructed URI, but keep the previous source + fallback, or merge nested + ontology-level lists.

3. Spanish comment — .cursor/05 §English-only artifacts (ontology-swrl.js)

// Raw editor sync — solo si NO estamos en modo raw; ...

Rewrite that comment in English. Product UI copy can stay localised; comments cannot.

4. No tests for new behaviour (.cursor/08, src/.coding_rules.md §10 / §11.7)

This is user-visible JS with a real behavioural fork (_dataPropsForClass, _currentUri, SWRL connected-set, raw-mode guard). There is already a pattern of source-level pytest under tests/units/front/ (see test_manual_mapping_panel_host.py). Please add at least:

• _currentUri: hash base, slash base, missing name, missing base
• _dataPropsForClass: nested props vs ontology-level domain fallback
• SWRL: raw-mode does not overwrite #swrlRawAntecedent / #swrlRawConsequent; inheritance one-hop inclusion
• Cohort template: #cohortDeleteBtn exists and is disabled until a rule is loaded

5. Changelog (.cursorrules post-change routine)

No /changelogs/v0.8.0/ section. Add one in English (title, context, numbered file list, tests).

Should-fix before merge

• CLA status is still pending on the PR (cla-assistant). We cannot merge until that is green — recheck if you already signed.
• Delete UX: deleteRule already uses native confirm(). .cursor/05 §Frontend forbids alert/confirm/prompt; use showDeleteConfirm from front/static/global/js/utils.js. Worth fixing on this path now that Delete is a first-class designer action.
• Axioms populateSelect: preferring name over uri is correct for in-domain classes and matches _resolve_uri. It will rewrite imported / foreign-namespace terms into the current base. If Expressions & Axioms can reference external IRIs, keep uri when it is not under the current base_uri.
• SWRL inheritance: only direct subclasses of a connected parent are added (one pass). Call that out in the comment, or walk the subclass closure if grandchild types should also be legal in a rule.
• CDN: jsdelivr for D3 matches the companion frontend PR — good, keep it identical (version pin). Don't land this file's CDN change if that companion is still churning the same URL.

Process / non-blocking

• Three independent fixes in one commit (Cohort delete, Cohort stale URI, SWRL/axioms). Fine if we want them in 0.8.0 together; painful to revert. Prefer two commits (Cohort vs SWRL/axioms) next time.
• Inline onclick="CohortModule.deleteActiveRule()" matches the existing New/Save buttons on this partial — not a new violation of .cursor/05 §Frontend (no new inline JS pattern).
• btn-outline-danger is fine: .cursor/11 greys Bootstrap colour variants on purpose.
• No session-data fields involved.
• GitHub reports no CI checks on the head branch. Confirm Actions actually ran on the fork PR.

What looks good

• Wiring delete through the existing DELETE /dtwin/cohorts/rules/:id instead of a new endpoint.
• _syncDeleteBtn gated on activeRuleId.
• Suppressing visual→raw overwrite while rawMode is on — that's the right source-of-truth split.
• Including direct subclasses of connected entities in the SWRL graph — reasonable UX, once documented.

Suggested retest (after the URI join fix)

1. Delete rule from the designer — confirm + list refresh + button disabled.
2. Slash base_uri (http://example.org/ns/) and hash base_uri — dropdown values must match ensure_uris.
3. Rebase domain, reload a previously saved rule — hops/filters should remap, not clear.
4. Axiom dropdowns: in-domain class after rebase; if you have an imported class, its IRI must not be rewritten.
5. SWRL: subtype with no own relationships is selectable; raw editor keeps typed atoms that are not graph nodes.

Happy to re-review once _currentUri matches _resolve_uri, _dataPropsForClass keeps a fallback, the Spanish comment is gone, and there is a changelog + a couple of front unit tests.