Design Direction
OntoGraph's design direction is to use formal ontology as a structuring filter for messy source material. Notes, excerpts, and imported documents can be captured without pretending that their wording is already a correct model. The system can then make uncertainty inspectable, ask for the missing information, and admit only explicit, evidence-bound structure through the governed write path.
This section is deliberately future-facing. It connects working components to the next product steps without presenting a proposal as shipped behavior.
Status legend
| Label | Meaning |
|---|---|
| Implemented | The current Rust core has executable tests for this bounded behavior. |
| Bounded proof | A separate tool or fixture proves part of the idea, but the end-to-end OntoGraph product flow is not available. |
| Planned | This is a design constraint or target architecture that still needs a spec, implementation, and executable acceptance evidence. |
In particular, the current core implements typed ontological validation, derived validation and search sidecars, and recoverable memory/file writes. A standalone reference-document adapter proves controlled Markdown parsing, exact UTF-8 byte spans, source-local candidates, and byte-exact reconstruction. Recursive folder ingestion, proximity-based fuzzy links, and a general gap-driven dialogue interface remain planned.
The proposed knowledge-refinement loop
raw note bytes
-> Source + exact Artifact identity
-> addressable document structure and source spans
-> source-local Assertion candidates + first-class Evidence
-> typed gaps and non-canonical candidate links
-> source-backed questions for a human
-> explicit OntoUML declarations in a proposed ChangeSet
-> pure candidate-state validation
-> Review -> Decision -> Promotion
No arrow in this diagram silently supplies the next one. Parsing is not interpretation. Co-occurrence is not Evidence. Validation is not approval. An approving Decision is not the Promotion event that makes a change current.
Articles
| Article | Central question |
|---|---|
| The Anatomy of an Unstructured Note Ingestion | How do exact bytes become source records, candidates, and span-grounded Evidence? |
| Formalizing the Unknown: A Topology of Gaps | How does the graph represent missing meaning, missing provenance, and explicit contradiction? |
| Calculus of Fuzzy Transitions | When may a contextual association advance toward an explicit OntoUML relation? |
| The Interactive Prompt Engine | How can gaps drive precise questions without letting questions become assertions? |
| Architectural Safeguards Against Oracle Hallucination | Which boundaries keep speculative reasoning outside accepted canonical state? |
The working thesis
The central claim is not that OntoUML can make uncertain text automatically true. It is that OntoUML supplies explicit obligations against which a proposed interpretation can be measured. The unmet obligations become useful data: missing identity roots, incomplete mediation, conflicting phases, missing source grounding, and unresolved interpretation.
That turns ingestion from a one-time cleanup step into a governed refinement cycle. OntoGraph can preserve a high-assurance core while retaining an equally inspectable perimeter of what remains unknown.
What this direction does not claim
These articles do not claim general natural-language understanding, arbitrary folder support, automatic ontology discovery, comprehensive OntoUML or gUFO conformance, production-backend equivalence, or autonomous promotion. See the published limitations for the current product boundary.