Skip to content

v1.8.0

Choose a tag to compare

@ThHanke ThHanke released this 11 Nov 18:12
· 33 commits to main since this release
bf0a5f3

Ontosphere v1.8.0

Reasoning moves to rdf-reasoner-konclude 0.9.3, repair suggestions are now computed by the reasoner, and large graphs reason and render much faster. Inferred counts change and some MCP behaviour changes, see Upgrade notes before updating scripts or agents.

Highlights

  • Faster reasoning on large graphs: runReasoning returns when the reasoner has finished. LUBM-1 with data: 126 s → 11.4–11.7 s (reasoner 8.6–8.8 s). The inferred triples then reach the canvas in chunks.
  • Faster, lighter canvas: elements are created in chunks and the editing buttons are mounted only for the hovered, selected or edited element. At 8,000 individuals the load settles in 19 s instead of 31 s. Renderer memory per individual drops from about 0.34 to 0.26 MB on large canvases (headless Chromium, 1 element + 2 links).
  • Repair and weakening run on the reasoner (repair(), weaken(), assessEdit()). explainDiagnostics on a large inconsistent ontology (PMDCO plus one clash) takes 128 s instead of 513 s.
  • More reliable startup: MCP tools answer only once startup loading has finished (window.__startupComplete()).

Upgrade notes

  • Inferred counts change with the reasoner. Pizza: 137 inferred triples with 0.8.0, 170 with 0.9.3. PMDco classification regains two direct edges (PMD_0020243 ⊑ PMD_0023003, PMD_0080022 ⊑ PMD_0023003). GALEN is unchanged (1,386). Re-measure any figure you published with 1.7.x.
  • MCP tool calls are rejected until startup loading has finished. Calls return success: false ("…still loading…") until window.__startupComplete() is true. Scripts and agents that call a tool right after the page opens must wait for that flag or retry.
  • An empty SHACL shapes URL now means "use the built-in ontology-quality shapes" (it used to mean "do not load shapes"). Turn SHACL off with the SHACL toggle.
  • explainDiagnostics no longer returns laconic explanations. laconicJustifications and the "Precise culprit" lines in repairBrief are gone; read justifications instead. Repair suggestions are unaffected.
  • explainDiagnostics justifications come from the reasoner's repair search: declarations are not part of a justification, and the number of justifications can exceed maxJustifications. Weakenings are offered only for a single independent clash.
  • Text results name IRIs as prefixed names or full IRIs, never as bare local names (inc:frank, not frank). Agents that parse these texts need to adapt.

Features and improvements

Reasoning

  • Repair and weakening on the reasoner; deletion repairs carry kind: "delete".
  • One explain per graph state, cached and invalidated on every write to the store.
  • explainDiagnostics skips unsatisfiableClasses for an inconsistent ontology (the list is meaningless there and cost 52 s on PMDCO); the output says unsatisfiableClassesSkipped: "ontology-inconsistent".
  • explainDiagnostics({ includeTimings: true }) returns per-stage timings.
  • The reasoner reads the real store: no copy, no write-back, inferences unchanged (about 1.4 s less on LUBM-1).

Canvas

  • Large canvases (more than 300 links): relation labels show only on selection. Unselected links are drawn as a bare line with arrowhead; they stay clickable, and selecting one brings back its label.
  • Entity cards use shared CSS classes (one DOM node fewer per card).
  • The halo's delete button (×) is gone; it duplicated the inline Delete. The Delete key and inline buttons work as before.

Startup and loading

  • window.__startupComplete(); shapes precedence: ?shaclShapes=, then the configured URL, then the built-in shapes.
  • ?workflowCatalog=off skips the PyodideSemanticWorkflow catalog load (headless runs, evaluations).
  • Bundled rdf, rdfs, owl and prov vocabularies are used when w3.org cannot be reached.

Fixes

  • Dataset and Turtle export no longer drop literals that differ only by language or datatype (PMDco lost 7 triples this way).
  • A stalled or crashed reasoner is a failed run (success: false, SHACL skipped) instead of a success.
  • explainEntailment resolves prefixed names; before, inc:frank was answered "NOT entailed".
  • SHACL violations without a focus node carry the shape as node id.
  • Blank-node values after reasoning conform again (fixed in reasoner 0.9.1).

Performance (8 cores, headless Chromium, reasoner 0.9.3)

reasoning-demo 1.1 s · Pizza 1.3–1.4 s · PMDCO 14.8–15.3 s · GALEN 3.9–4.0 s · LUBM-1 with data 11.4–11.7 s. Earlier LUBM-1 and GALEN figures in the README loaded an OWL/XML TBox the RDF parser cannot read, so no inference took place; they are replaced. Memory: LUBM-1 through a ?rdfUrl= load peaks at 3.4 GB renderer memory in headless Chromium.

The full list with commits is in CHANGELOG.md.

What's Changed

Full Changelog: v1.7.7...v1.8.0