Skip to content

LICENSING.md

Latest commit

 

History

History
308 lines (250 loc) · 18.1 KB

LICENSING.md

File metadata and controls

308 lines (250 loc) · 18.1 KB
Raw

Licensing

Authoritative scope declaration for this repository. Where any other file in this repository disagrees with this one about which licence covers which path, this file governs.

Copyright (c) 2024–2026 Dr John O'Hare / DreamLab AI Consulting Ltd, except for the third-party material identified in NOTICE.

This repository is not under a single licence. It contains three kinds of thing: original software, a vendored MIT-licensed derivative, and a machine-generated database. Each is licensed on its own terms.

Scope table

Path Licence SPDX Full text
pipeline/ GNU Affero General Public License v3.0 or later AGPL-3.0-or-later LICENSE
.github/ (repository root only) GNU Affero General Public License v3.0 or later AGPL-3.0-or-later LICENSE
static/ GNU Affero General Public License v3.0 or later AGPL-3.0-or-later LICENSE
docs/ (original prose) GNU Affero General Public License v3.0 or later AGPL-3.0-or-later LICENSE
examples/ GNU Affero General Public License v3.0 or later AGPL-3.0-or-later LICENSE
README.md, LICENSING.md, COMMERCIAL.md, CONTRIBUTING.md, NOTICE GNU Affero General Public License v3.0 or later AGPL-3.0-or-later LICENSE
explorer/ MIT License MIT LICENSE-EXPLORER
ontology/ Open Database License v1.0 ODbL-1.0 LICENSE-DATA
dist/ Open Database License v1.0 ODbL-1.0 LICENSE-DATA

LICENSE at the repository root is the AGPL-3.0 text. GitHub's licence detector reads that file and will label the whole repository "AGPL-3.0". That label is wrong for explorer/ and ontology/ and dist/. This table is the correct statement.

The .github/ row covers the repository-root workflow directory only. The explorer carries its own explorer/.github/workflows/wasm-publish.yml; that file is inside explorer/ and is MIT, like the rest of that tree.

Paths not listed (CNAME, .gitignore) are configuration with no meaningful authorship and are not claimed. dist-ci/ is not a repository path either: it exists only inside a CI run, as the build target and upload artefact of .github/workflows/build.yml, and is never committed.

What is in each path

File counts are of files Git will track: git ls-files --cached --others --exclude-standard <path>. The node_modules/, target/, pkg/, __pycache__/ and dist/api/ trees excluded by .gitignore are excluded here too. Verified against the working tree on 2026-07-25.

  • pipeline/: 13 files. Python 3 / rdflib. Seven stages: parse, validate, Turtle, WebVOWL JSON, Page API, search index, graph tiers (NGG1 binary). Entry points are pipeline.build, pipeline.validate, pipeline.jsonld_to_turtle. Wholly original work; no vendored code.
  • static/: ns/v2.jsonld, the JSON-LD @context document that maps the corpus vocabulary onto owl:, rdfs:, skos:, prov: and the local vc: namespace (https://narrativegoldmine.com/ns/v1#). Original work.
  • explorer/: WasmVOWL. Subtrees: modern/ (React + TypeScript SPA), rust-wasm/ (Rust → WebAssembly layout engine and NGG1 reader), docs/, tests/, scripts/, .github/workflows/wasm-publish.yml, plus the design and format notes at the tree root (FORMAT-NGG1.md, VOWL-SPEC.md, CAPABILITIES.md and others) and license.txt. A derivative of WebVOWL. See "Why the explorer is MIT" below. No file count is stated here because the tree is under active development; the licence covers the whole subtree regardless of its size.
  • ontology/: pages/, 7457 Markdown files and nothing else, each a Logseq page carrying an embedded JSON-LD block. This is the corpus. 7457 is a contract, not a statistic: .github/workflows/build.yml gate 3 fails the build if the compiled class count is anything other than EXPECTED_CLASSES: '7457'.
  • docs/, examples/: original prose. examples/ is present in the working tree but currently holds no files; the AGPL row above covers it if and when it is populated, and licenses nothing today.
  • dist/: the built dataset, 11 files. dist/data/ holds ontology.ttl (12,301,363 bytes of Turtle, 252,974 triples), ontology.json (39,337,821 bytes of WebVOWL) and dist/data/graph/ (seven NGG1 binary tiers: full.bin and one per domain, plus overview.json and stats.json). These are derived from ontology/ and carry the same licence as their source. dist/api/ is not committed (see .gitignore); it is rebuilt by python -m pipeline.build ontology/pages dist.

Why the software is AGPL-3.0-or-later

pipeline/ is wholly original. The AGPL's network clause (§13) is the point: this is build-and-publish machinery whose natural deployment is as a service that never distributes a binary. Under GPL-3.0 a hosted derivative would carry no obligation to release corresponding source. Under AGPL-3.0 it does.

"or later" is deliberate. It permits users to move to a future AGPL version without seeking permission.

Why the corpus is ODbL-1.0

Rights being licensed

ODbL-1.0 is a database licence. It grants share-alike terms over three things at once, which matters here because at least one of them is contested:

  1. Copyright in the database as a compilation: the selection and arrangement of the 7457 pages, the class hierarchy, the relation vocabulary.
  2. Sui generis database right under the Copyright and Rights in Databases Regulations 1997 (SI 1997/3032), which subsists where there has been substantial investment in obtaining, verifying or presenting the contents. Term: 15 years from the end of the year of making, or of first publication if within that period (reg. 17).
  3. Contract, for users who accept the licence, independent of whether any of the above subsists.

Contents grant. The ODbL preamble states that ODbL "only governs the rights over the Database, and not the contents of the Database individually", and advises licensors to pair it with a contents licence. DreamLab AI additionally and expressly licenses the individual contents of ontology/ and dist/ (the page prose, definitions, labels and axioms) on the same ODbL-1.0 terms. Attribution and share-alike therefore apply to extracts as well as to the database as a whole.

Rights basis: UK CDPA 1988 s.9(3)

The corpus is mostly AI-generated. Under UK law that is not fatal to authorship.

CDPA 1988 s.9(3): "In the case of a literary, dramatic, musical or artistic work which is computer-generated, the author is the person by whom the arrangements necessary for the creation of the work are undertaken."

CDPA 1988 s.178: "'computer-generated', in relation to a work, means that the work is generated by computer in circumstances such that there is no human author of the work."

The arrangements necessary for the creation of this corpus (the schema design, the prompt and generation harness, the domain and category taxonomy, the validation gates, the selection of which pages are published) were undertaken by DreamLab AI. On that basis DreamLab AI is the author under s.9(3) and licenses accordingly. Duration for a computer-generated work is fifty years from the end of the calendar year in which the work was made (s.12(7)), shorter than the life-plus-70 term for human-authored works.

Position differs by jurisdiction

s.9(3) has no US equivalent, and US law points the other way. In Thaler v. Perlmutter, No. 23-5233 (D.C. Cir., decided 18 March 2025), the Court of Appeals affirmed the refusal of registration, holding that "the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being". The court expressly declined to reach Thaler's separate argument that he was the author by virtue of making and using the generating machine, so the precise treatment of a "made-the-arrangements" claimant remains open in the US, but the human-authorship floor is settled at the D.C. Circuit.

The practical consequence: the copyright layer of the ODbL grant may be worth little or nothing in the United States. What survives there is the contract layer: ODbL is drafted as an agreement, and its terms bind a user who accepts them regardless of whether copyright subsists. What does not survive is the sui generis database right, which has no US analogue at all.

The EEA position differs again. Following the Intellectual Property (Copyright and Related Rights) (Amendment) (EU Exit) Regulations 2019, UK citizens, residents and businesses are not eligible to hold sui generis database right in the EEA for databases created on or after 1 January 2021. This database was created after that date by a UK company. Its sui generis database right therefore subsists in the UK only.

s.9(3) is itself under review. The UK IPO's 2021–22 consultation on AI and IP concluded to make no change to s.9(3). The December 2024 "Copyright and AI" consultation reopened it, and the subsequent government Report on Copyright and Artificial Intelligence records both majority respondent support for removing s.9(3) and "minimal evidence that CGWs protection is being used". Removal is a live policy option. s.9(3) remains in force as at the date of this document. If it is repealed, the copyright basis for the ODbL grant weakens in the UK too, and the contract and database-right layers carry more of the load.

None of this is hedging for its own sake. It is the actual state of the law, and anyone relying on the corpus should know which layer of the grant is doing the work in their jurisdiction.

What the corpus is

Mostly AI-generated synthetic content, produced under human direction, by design. It is an ontology testbed built to exercise the pipeline at medium scale: 7457 OWL classes, 110,617 declared edges of which 96,377 resolve and are emitted, 252,974 triples. It is not an authoritative encyclopaedia and should not be cited as one.

The provenance metadata attests traceable generation under human direction, not human authorship. Do not read it as the latter. It lives in two places, and they differ:

  • the JSON-LD blocks under ontology/pages/ carry did:nostr identifiers, URNs and prov:generatedAtTime timestamps, though not on every page: at 2026-07-25, 6169 of the 7457 pages carry a did:nostr and 3966 carry a prov:generatedAtTime;
  • dist/data/graph/stats.json carries attributedTo: "did:nostr:jjohare", provenance.did, and corpus.nature: "synthetic". The pipeline emits the framing as data rather than leaving it to prose.

dist/data/ontology.ttl carries none of it: the Turtle writer emits no prov: terms at all. If you are reading the corpus as Turtle, the provenance is in stats.json and in the source pages, not in the triples.

Why the explorer is MIT

explorer/ is a derivative of WebVOWL, copyright (c) 2014–2019 Vincent Link, Steffen Lohmann, Eduard Marbach, Stefan Negru and Vitalis Wiens, licensed MIT. The upstream is VisualDataWeb/WebVOWL; the immediate parent is DreamLab-AI/WasmVOWL (branch master), an MIT fork of it.

MIT permits sublicensing. DreamLab AI could relicense this derivative under AGPL-3.0-or-later, and an earlier draft of NOTICE said it had. That has been reversed. The reason is that the relicence would be hollow: substantially identical code is published under MIT one repository away at DreamLab-AI/WasmVOWL. Anyone wanting the explorer without copyleft obligations can take the MIT copy and reimplement the delta. An AGPL claim over explorer/ would impose obligations on honest readers of this repository while restricting nobody who cared to look, so it would raise the cost of compliance without raising the cost of appropriation. It is not asserted.

The MIT grant is therefore passed through unchanged. LICENSE-EXPLORER is a byte-exact copy of explorer/license.txt, which is itself the upstream text. The copyright notice above must be preserved in all copies and substantial portions.

Internal manifests corroborate this: explorer/package.json declares "license": "MIT" and explorer/rust-wasm/Cargo.toml declares license = "MIT". explorer/modern/package.json carries no license field; it is covered by this declaration and by LICENSE-EXPLORER.

explorer/ is not a verbatim copy of WasmVOWL: the upstream legacy/ tree (the original D3 WebVOWL implementation) is not vendored here.

Third-party runtime dependencies of the explorer are declared, not vendored, in explorer/package.json, explorer/modern/package.json and explorer/rust-wasm/Cargo.toml. Each remains under its own licence, held by its own copyright holders. Nothing in this document supersedes those licences.

Licence compatibility

The three licences coexist because they cover disjoint paths and the combined work is not a single program:

  • pipeline/ (AGPL) reads ontology/ (ODbL) and writes dist/ (ODbL). Reading and writing data does not make the data a part of the program, nor the program a derivative of the data. The ODbL's share-alike attaches to derived databases, not to the software that derives them.
  • explorer/ (MIT) loads dist/data/* (ODbL) at runtime over HTTP. MIT code is compatible with being distributed alongside AGPL code; they are separate works aggregated in one repository, not linked into one binary.
  • MIT is one-way compatible with AGPL-3.0. If you combine explorer/ with pipeline/ into a single derived program, the result is AGPL-3.0-or-later and the MIT notice must be preserved. The reverse combination is not permitted.

If you redistribute only part of this repository, carry the licence for that part, and carry NOTICE.

Text and data mining, and model training

Read this section before assuming a licence stops anyone training on this corpus. In most jurisdictions it does not.

  • EU. Directive (EU) 2019/790 Art. 4(1) permits reproductions for text and data mining of lawfully accessible works, for any purpose including commercial. Art. 4(3) makes that exception conditional: it applies "on condition that the use of works and other subject matter referred to in that paragraph has not been expressly reserved by their rightholders in an appropriate manner, such as machine-readable means in the case of content made publicly available online." Recital 18 confirms that for content made publicly available online, machine-readable means (including metadata and the terms and conditions of a website or service) are the appropriate way to reserve. A licence file in a Git repository is unlikely, on its own, to satisfy that standard.
  • UK. The exception in force is CDPA s.29A, and it is narrow: copies for computational analysis "for the sole purpose of research for a non-commercial purpose", by a person with lawful access. Commercial AI training is not covered by s.29A. Note s.29A(5): a contract term purporting to restrict copying that s.29A permits is unenforceable, so the ODbL cannot override the non-commercial research exception. A broader "any purpose" exception was announced in the 2022 government response and was not enacted. The December 2024 "Copyright and AI" consultation proposed an exception coupled with a rights-reservation mechanism; the subsequent government report states that a broad exception with opt-out is no longer the preferred way forward. Nothing new is in force.
  • US. Fair use is asserted for training by most model developers and is being litigated. Combined with Thaler, the copyright layer over machine-generated corpus text is weak there in both directions.

Express reservation

DreamLab AI Consulting Ltd expressly reserves the rights of text and data mining, including the use of any part of ontology/ or dist/ for training, fine-tuning, evaluating or otherwise developing machine-learning models, under Article 4(3) of Directive (EU) 2019/790 and under any equivalent national provision. This reservation applies to the whole of the database and to any substantial part of it, and to every distribution channel through which it is made available, including https://narrativegoldmine.com/ and this repository. It does not restrict acts permitted by CDPA s.29A.

Licences for such use are available. See COMMERCIAL.md.

Current machine-readable coverage. The reservation above is carried in this file and is intended to be quotable verbatim. As at 2026-07-25 the published site serves no robots.txt (https://narrativegoldmine.com/robots.txt returns HTTP 404 with the SPA fallback page, content-type: text/html, not a robots file), no TDM-Reservation-Path header on https://narrativegoldmine.com/, and no /.well-known/tdmrep.json (also 404). The emitted dist/data/ontology.ttl binds http://purl.org/dc/terms/ to the prefix dc1: and uses it for dc1:creator only: there is no dc1:license and no dc1:rights triple in the file. Those are the signals a machine-readable reservation under Art. 4(3) would normally need, and they are absent. Anyone relying on the absence of a signal should read this section instead; anyone auditing the reservation should treat the missing signals as a known gap in this repository rather than as evidence that no reservation was made.

Contact and alternative terms

The split above is what is granted publicly. Terms other than AGPL-3.0 for the pipeline, and terms other than ODbL share-alike for the corpus, are available by negotiation. See COMMERCIAL.md.

Contributions are accepted only on terms that keep alternative licensing possible. See CONTRIBUTING.md.

Disclaimer

This document explains the licensing position DreamLab AI has adopted and the reasoning behind it. It is not legal advice, and it is not a warranty that any particular right subsists in any particular jurisdiction: the sections above say plainly where it may not. The operative grants are the licence texts in LICENSE, LICENSE-DATA and LICENSE-EXPLORER, each of which disclaims warranties in its own terms. If you need certainty for your use, take your own advice or ask for a written licence.