Licences and paradata¶
This page explains, without code, what the ATRIUM tools record about the software, models and data they use, what the one licence they compute means and does not mean, and what replaces it.
What a run records¶
Every run of an ATRIUM tool writes a short log, its paradata. It says which program and version ran, when, with which settings, and what it produced. It also lists the components the run used: the models, datasets, vocabularies and web services. Each component is listed with its own licence, or, for a web service, the terms of use it is offered under.
The document record that travels through the pipeline points to the paradata of every run that wrote to it. So for any record you can find out which tools touched it and what each of them used.
The licence the tools compute, and what it is not¶
From that list of components, the shared module para_licenses.py also computes one label: the most
restrictive of the component licences, on a scale from public domain to CC BY-NC-SA 4.0. It was added as a
warning. If a run used a model published for non-commercial use, the label says so at a glance. The record
carries the label as provenance.license, and an RO-Crate export states it as the licence of the crate.
The label is not the licence of the output, and should not be read as one:
- A licence covers what it is attached to: a program, a model or a dataset. Using a tool does not put what it produces under the tool's licence.
- Web services have terms of use, not a licence for their output. LINDAT, for example, licenses its software and models and offers its services under Terms of Use.
- "The strictest wins" says nothing about compatibility. Some licences cannot be combined at all. Picking the strictest one does not show that the others are respected.
- Using a component is not, by itself, making a derived work of it. For example, tagging a line with one term from a vocabulary. Whether an output is a derived work is a legal question, and the label cannot answer it.
- The rights in an output depend first on the source document. No tool knows those rights.
The tools' pages now list each component with its licence or terms. They do not say that a result "is" under any licence.
What changes¶
The proposal on #6 keeps what is useful and drops the label:
- Keep the list of components, each with its own licence or terms and where that was read, and make it more precise.
- Record which processing policy allowed the run. The deployment's operator (AMČR, for the pilot) writes the policy: for example, whether texts may be sent to an external service. Each document also carries its own permission, and a document without one is not sent to an external service.
- State the rights in an output only when someone entitled to do so asserts them. Without that, the rights are unknown, and the record says nothing.
AMČR agreed to this on 22 October 2026: AMČR states the rights of the outputs it publishes, and no tool makes
an automatic or default licence assertion. The shared modules change next, first where records and crates still
state the computed label. Until then, read provenance.license as "the most restrictive licence among the
components used", nothing more.
Where the details are¶
| To see | Read |
|---|---|
| each tool's components and their licences | the tool's workflow page, under Workflows |
| how the record grows and points to paradata | The document contract |
| every field of the paradata file | docs/paradata_schema.md |
| the proposal in full, for implementers | docs/proposals/rights-and-policy.md |
| the discussion | #6, #4, #78 |