Skip to content

ecosystem.md

Latest commit

 

History

History
311 lines (253 loc) · 15.3 KB

ecosystem.md

File metadata and controls

311 lines (253 loc) · 15.3 KB
Raw

Ecosystem

Where this repository sits among the DreamLab-AI projects, what consumes the corpus, and which neighbouring repository names are empty.

Every repository below was checked with gh api /repos/DreamLab-AI/<name> on 2026-07-25. Repositories that could not be verified, or that turned out to contain something other than their name suggests, are either excluded or called out explicitly under Reserved and empty names.

What this repository is

A Logseq-authored ontology corpus and the Python/rdflib pipeline that turns it into published RDF, a WebVOWL-format graph, a page API, a search index, and binary graph tiers.

Component Path Licence Licence file
Build pipeline (10 Python modules + tests) pipeline/ AGPL-3.0-or-later LICENSE
Corpus (7,874 public pages) ontology/pages/ ODbL-1.0 LICENSE-DATA
Explorer (WasmVOWL derivative) explorer/ MIT LICENSE-EXPLORER, explorer/license.txt
JSON-LD context document static/ns/v2.jsonld AGPL-3.0-or-later LICENSE
Built artefacts dist/data/, dist/api/ ODbL-1.0 (data) LICENSE-DATA

The 2026-07-25 build produced 7,874 OWL classes, 0 individuals, 7,870 pages, 98,776 emitted edges (of 111,827 declared) and 258,200 Turtle triples, in 18.3 seconds. The class, page, node, edge, domain (6) and category (34) counts, plus pipelineVersion: ng-1.0.0 and datasetDate: 2026-07-25, are recorded in dist/data/graph/stats.json. The triple count is a property of dist/data/ontology.ttl, not of stats.json; rdflib reports 252974 on parsing it. The 18.3 s elapsed time is printed by pipeline/build.py at the end of a run and is not persisted anywhere. stats.json also carries the corpus honesty statement "corpusNature": "synthetic-ai-generated-human-directed". The corpus is mostly AI-generated synthetic content produced under human direction, by design: an ontology testbed, not an authoritative encyclopaedia. Its provenance metadata (did:nostr, generatedAtTime, URNs) attests traceable generation under human direction, not human authorship.

Corpus flow

flowchart TD
    SRC["ontology/pages/<br/>7,874 Logseq .md pages<br/>with embedded JSON-LD"]

    subgraph PIPE["pipeline/ — Python + rdflib, AGPL-3.0-or-later"]
        P1["jsonld_parser.py"] --> P2["validate.py<br/>0 errors, 0 warnings, 1,402 info"]
        P2 --> P3["jsonld_to_turtle.py"]
        P2 --> P4["jsonld_to_webvowl.py"]
        P2 --> P5["jsonld_to_page_api.py"]
        P2 --> P6["jsonld_to_search.py"]
        P2 --> P7["emit_graph_tiers.py<br/>NGG1 binary"]
    end

    SRC --> P1

    TTL["dist/data/ontology.ttl<br/>258,200 triples · 12.0 MiB"]
    VOWL["dist/data/ontology.json<br/>WebVOWL · 37.5 MiB"]
    TIERS["dist/data/graph/*.bin<br/>full + 6 domain tiers<br/>+ overview.json + stats.json"]
    API["dist/api/pages/*.json + markdown/*.md<br/>+ dist/api/search-index.json"]

    P3 --> TTL
    P4 --> VOWL
    P7 --> TIERS
    P5 --> API
    P6 --> API

    subgraph SITE["narrativegoldmine.com — GitHub Pages, branch gh-pages"]
        EXP["explorer/ (WasmVOWL)<br/>React+TS SPA · Rust→WASM layout<br/>NGG1 tier reader · MIT"]
        CTX["/ns/v2.jsonld<br/>JSON-LD context"]
    end

    TIERS --> EXP
    VOWL --> EXP
    API --> SITE
    TTL --> SITE
    CTX -.->|"@context of 7,531 JSON-LD blocks"| SRC

    subgraph VC["DreamLab-AI/VisionClaw — AGPL-3.0"]
        SYNC["GitHubSyncService<br/>fetch markdown + ontology"]
        OXI["Oxigraph<br/>embedded RocksDB RDF quad store"]
        WHELK["Whelk-rs<br/>OWL 2 EL reasoner"]
        SYNC -->|"write asserted triples<br/>(SHACL-gated)"| OXI
        OXI --> WHELK
        WHELK -->|"inferred axioms + PROV-O"| OXI
    end

    TTL --> SYNC
    SRC --> SYNC
Loading

One caveat on that context edge: static/ holds exactly one file, static/ns/v2.jsonld, and it is the only context document the deploy publishes. The corpus does not reference it uniformly. Counting "@context" values across ontology/pages/*.md gives 7,531 pointing at /ns/v2.jsonld, 9,546 at /context/v1.jsonld, 1,538 at /ns/v1, 12 at //visionclaw.dreamlab-ai.systems/ns/v1 and 1 at //w3id.org/did/v1. Only the first of those dereferences on narrativegoldmine.com. This does not affect the build, since pipeline/jsonld_parser.py never fetches a remote context, but a consumer that expands the JSON-LD blocks over the network will fail on the majority of them.

The VisionClaw half of that diagram is transcribed from VisionClaw's own canonical architecture doc, docs/explanation/system-overview.md: the message sequence at lines 193-196 (GitHubSyncService → Oxigraph → Whelk reasoner), the store description at line 122 ("an embedded, RocksDB-backed RDF quad store opened in-process"), and the stack table's "Graph store" row at line 244.

Verified siblings

DreamLab-AI/VisionClaw: AGPL-3.0, branch main

Logseq Spring Thing Immersive & Agentic Knowledge Development Engine. The second consumer of this corpus. It fetches the markdown and the OWL from GitHub, gates writes with SHACL-lite, stores them in an embedded Oxigraph quad store, and reasons over them with Whelk-rs (OWL 2 EL). Where this repository publishes a static artefact, VisionClaw runs the same triples as a live, reasoned graph.

Its NOTICE (1,513 bytes) explains the AGPL-3.0 relicence: at milestone M1 of ADR-032 the webxr workspace began linking solid-pod-rs, solid-pod-rs-nostr and solid-pod-rs-idp, and "Linking these crates makes the combined webxr binary AGPL-3.0-only." LICENSE.MPL survives in the repo as a historical file; VisionClaw is not MPL 2.0 today, despite what VisionFlow's docs/architecture/licensing.md line 11 still says.

Useful entry points, all confirmed present on main: system-overview · insight-migration-loop · graph-schema · rest-api

DreamLab-AI/WasmVOWL: MIT, branch master

A fork of VisualDataWeb/WebVOWL, described as "Visualizing ontologies on the Web". This is the direct upstream of explorer/. Both carry the same MIT text: explorer/license.txt line 3 reads "Copyright (c) 2014-2019 Vincent Link, Steffen Lohmann, Eduard Marbach, Stefan Negru, Vitalis Wiens".

Two divergences worth knowing. The default branch is master, not main, and its newest commits sit on the unmerged fix/audit-2026-07-10 (36105cc3, 2026-07-10); master HEAD is 545885c8, "clean up directories", 2025-11-10. The two have diverged: the branch is 3 ahead and 1 behind master. And explorer/ here is not a mirror in either direction: upstream master carries legacy/, modern/, rust-wasm/, tests/, whereas explorer/ drops the legacy/ D3 tree and adds docs/ and scripts/, giving docs/, modern/, rust-wasm/, scripts/, tests/.

explorer/ stays MIT deliberately. AGPL-ing it would be hollow while the same WebVOWL-derived code is MIT one repository away: explorer/package.json and explorer/license.txt still hash to the identical git blobs as their upstream master counterparts (f01b936d and 6df36dbf).

DreamLab-AI/VisionFlow: no root licence, branch main

Distributed coordination platform — governed AI agent meshes with shared semantics, self-sovereign data, and human-in-the-loop judgment. The ecosystem's documentation and website canon; its GitHub homepage field is empty, so it publishes no canonical site URL. It carries no root LICENSE (HTTP 404); LICENSES/README.md states that "Until a root licence file is added, treat repository-local documentation and website assets as requiring explicit maintainer confirmation before reuse outside the DreamLab AI ecosystem." Its docs/architecture/licensing.md is dated 2026-05-20 and is stale on VisionClaw; cite VisionClaw's own LICENSE and NOTICE instead.

DreamLab-AI/agentbox: AGPL-3.0, branch main

NIX based agent container. The agent runtime this pipeline was built inside. Third-generation fork: agentbox ← VisionClaw ← origin-logseq-AR. No code or data dependency in either direction; it is tooling provenance, not a consumer.

DreamLab-AI/Metaverse-Ontology: no licence file, branch main

The closest independent precedent to this repository's method. Its README: "A formal ontology for metaverse concepts using an innovative hybrid approach that combines Logseq markdown for human readability with OWL Functional Syntax for formal reasoning", extracted by a Rust logseq-owl-extractor/.

The methods differ where it matters: Metaverse-Ontology reads Logseq properties and emits OWL Functional Syntax from Rust; this repository parses embedded JSON-LD blocks with Python/rdflib and emits Turtle, WebVOWL JSON and NGG1 binary tiers, set out in methodology/the-hybrid-approach.md. Its root tree carries no LICENSE file and GET /repos/DreamLab-AI/Metaverse-Ontology/license returns HTTP 404, so nothing there is safe to vendor.

Indirect: DreamLab-AI/solid-pod-rs, AGPL-3.0, branch main

Rust-native Solid Pod server (LDP + WAC + NIP-98 + OIDC + Notifications). No direct relationship here; listed because, per VisionClaw's NOTICE, it is the cause of the licence that governs the downstream consumer of this corpus.

Reserved and empty names

Two repository names in the organisation describe work this repository now does. Neither contains code. Stating this plainly is cheaper than leaving readers to discover it.

  • DreamLab-AI/logseq-ontology-toolkit: MIT, branch main, described as "automation for hybrid logseq ontology and web presentation". Its entire git tree is one blob: LICENSE, 1,083 bytes. One commit, 4b8cc5ed "Initial commit", 2025-11-13. Its stated purpose is precisely what pipeline/ and explorer/ implement here.
  • DreamLab-AI/KG-markdown: no licence, no description, no commits. Both /git/trees/HEAD and /commits return HTTP 409 "Git Repository is empty."

Both were registered in November 2025 for work that subsequently landed here. The right resolution is upstream: archive them, or push a one-line README pointing at DreamLab-AI/knowledgeGraph, rather than annotating the ambiguity forever.

Excluded deliberately: DreamLab-AI/VisionFlow-Ontology-Engine is not an ontology engine. It is a fork of ShinMegamiBoson/OpenPlanter whose README opens "# OpenPlanter — A recursive-language-model investigation agent with a terminal UI, built in Rust" and whose crates are op-cli, op-core, op-engine, op-model, op-runtime, op-scripts, op-tools, op-tui. Nothing in it relates to this corpus.

Publication topology

The corpus is authored and built in a private source repository (jjohare/logseq) and deployed cross-repo into this one. From .github/workflows/publish.yml in that source repo:

- name: Deploy to GitHub Pages
  uses: peaceiris/actions-gh-pages@v3
  with:
    personal_token: ${{ secrets.ACCESS_TOKEN }}
    external_repository: DreamLab-AI/knowledgeGraph
    publish_branch: gh-pages
    publish_dir: www
    commit_message: "Deploy JSON-LD v3 pipeline from jjohare/logseq @ ${{ github.sha }}"

The same job writes narrativegoldmine.com into www/CNAME, matching this repo's root CNAME. GitHub's Pages API reports html_url https://narrativegoldmine.com/, source branch gh-pages, cname narrativegoldmine.com, status built. Published endpoints, as enumerated by the workflow's own deployment summary:

  • https://narrativegoldmine.com/: explorer
  • https://narrativegoldmine.com/ns/v2.jsonld: JSON-LD context
  • https://narrativegoldmine.com/data/ontology.ttl: Turtle
  • https://narrativegoldmine.com/data/ontology.json: WebVOWL graph
  • https://narrativegoldmine.com/api/search-index.json: search index

A markdown-mirror gate in the same job fails the build if the count of copied public pages differs from the parser's vc:public count, so the published page API cannot silently drift from the corpus.

This repository's own workflow, .github/workflows/build.yml, is the reproducible half of that job and deliberately has no deploy path: its whole permission set is contents: read, so it could not push even if a step tried. It runs a secret scan over ontology/, the pipeline unit tests (pipeline/tests, including the 183-byte NGG1 golden fixture), the 7-stage build into dist-ci/, a hard contract gate asserting EXPECTED_CLASSES: '7874' against both stats.json and ontology.json, and python -m pipeline.validate ontology/pages, which must report zero errors. It now also reports zero warnings. The 1,402 remaining issues are MULTI_PARENT at info severity: bridging classes, which are a design property of this corpus rather than a defect. They were published as warnings until 2026-07-25, when the count was corrected. See dist/api/validation-report.json. Reproduce it with:

pip install "rdflib>=7.0.0" pytest
python -m pytest pipeline/tests -q
python -m pipeline.build ontology/pages dist-ci

Cross-referencing this repository from a sibling

Maintainers of the sibling projects can paste the following into their README. It states only claims verifiable from files in this repository.

### Ontology corpus

The ontology this project consumes is published from
[DreamLab-AI/knowledgeGraph](https://github.com/DreamLab-AI/knowledgeGraph)
and served at [narrativegoldmine.com](https://narrativegoldmine.com).

- Turtle: <https://narrativegoldmine.com/data/ontology.ttl> — 7,874 OWL classes,
  258,200 triples, 6 domains, 34 categories (build of 2026-07-25).
- WebVOWL graph: <https://narrativegoldmine.com/data/ontology.json>
- JSON-LD context: <https://narrativegoldmine.com/ns/v2.jsonld>
- Search index: <https://narrativegoldmine.com/api/search-index.json>
- Build statistics and provenance:
  <https://github.com/DreamLab-AI/knowledgeGraph/blob/main/dist/data/graph/stats.json>

Licensing: the corpus is ODbL-1.0; the Python/rdflib pipeline is
AGPL-3.0-or-later; the `explorer/` visualiser is MIT, being a derivative of
[WebVOWL](https://github.com/VisualDataWeb/WebVOWL) via
[DreamLab-AI/WasmVOWL](https://github.com/DreamLab-AI/WasmVOWL) (branch
`master`).

The corpus is mostly AI-generated synthetic content produced under human
direction, by design. It is an ontology testbed, not an authoritative
encyclopaedia; its provenance metadata attests traceable generation under human
direction, not human authorship.

Two smaller edits worth making at the same time:

  1. In VisionFlow, docs/architecture/licensing.md line 11 records VisionClaw as MPL 2.0. VisionClaw's root LICENSE is AGPL-3.0 and its NOTICE gives the reason. Correct the row rather than propagating it.
  2. In logseq-ontology-toolkit and KG-markdown, either archive or add a README redirecting to DreamLab-AI/knowledgeGraph.