When a dataset's licence does not answer the question: a disagreement catalogue
Ambiguity, not prohibition, is the common case. Six shapes of ambiguity that block reuse, four of which have no observed instance yet and say so.
Verdict
SIX OF NINE CANDIDATES CARRIED A LICENCE AMBIGUITY
Across nine independently examined candidates, only two distinct disagreement shapes occur: no licence claim at artifact scope, and a non-licence label sitting in the licence slot.
Neither is a prohibition. In every case the artifact was downloadable, and in none of them did the record establish that it could be reused.
Pattern 1: no licence claim covers the artifact
Three candidates produced zero licence claims. Their landing pages and file metadata carry no reuse licence at all, so the rights stage has nothing to quote and the verdict is itself the finding.
Two are thermochemical property records; the third is an atmospheric dispersion dataset. All three are freely downloadable.
The recorded open question is the same in each case: which licence applies to the artifact? No applicable claim was found, and a public download is not reuse permission.
This is also the reason these three must never be made to produce claims. The archives in question publish no reuse licence, so a claim set would be invented evidence rather than discovery.
Pattern 2: a label that is not a licence, in the place a licence goes
Six of the nine candidates carry a label in the licence slot that grants nothing: a repository status of other-open, a bare copyright notice, an access statement, or a publisher terms URL that resolves to no licence.
The pattern appears on both sides of the verdict line, and the distinction is worth stating precisely. The label alone is never a veto. It decides the verdict only when it is the only applicable claim.
| Situation | Verdict | What the record actually said |
|---|---|---|
| An access statement is the only claim | Unclear | “All information can be accessed by everybody.” Access is not permission. |
| A repository status label is the only claim | Unclear | A licence id of other-open, which names no terms. |
| A real licence also applies | Reusable | The label is recorded and shown, but absence does not conflict with a grant. |
The access-statement case is instructive. The claim is examined and recorded rather than ignored, so the verdict rests on evidence rather than on an empty claim set. It still grants no reuse.
The case that cost a verdict rule: a permissive grant beside an unresolved deferral
One candidate, a battery degradation dataset, emits three claims in tension. The record carries a status of other-license-specified and a government-works pointer. The bundled archive carries an MIT licence file.
The MIT grant covers, in its own words, the software. Whether it reaches the data is a separate question that the record does not answer, and the two record labels defer to terms they never name.
The governing rule is that a permissive grant never silently outranks an unresolved deferral. The verdict is unclear, and the open question is recorded: establish what the artifact itself is released under.
This is the defect that justifies freezing the independent verdict set. An earlier implementation let the permissive claim win. That bug passed 178 unit tests and was caught only because the independent verdicts disagreed with it.
Four patterns with no observed instance
The remaining entries are named, defined, and empty. Each describes a shape that the resolution rules handle and that has not yet appeared across the nine candidates examined.
Reporting them as empty is a deliberate choice. An earlier attempt to stretch one observed instance into a second pattern, so that every name would point at a real candidate, was rejected in review.
Four honestly empty entries are the correct outcome. The alternative buys apparent completeness at the cost of the one property that makes a catalogue worth consulting.
How these observations were established
| Candidate set | Nine, with verdicts established independently in a previous round |
| Source of quotations | Verbatim from the rights output of a single dated run, one record per candidate |
| Disagreement kinds observed | Scope gap on an empty claim set; non-licence label |
| Patterns exercised by tests | Pattern 2, by a repository-status case and a marker-beside-grant case |
| Verdict determinism | Rules are tested; conflicts are reported with claim ids and never auto-resolved |
Every statement quoted in this catalogue is verbatim from the rights output of that run rather than paraphrased from a landing page. The distinction matters: a paraphrase of a licence is not evidence about a licence.
What this means
The common obstacle to reusing public data is not a refusal. It is a record that never addressed reuse at all, and the correct response is to name the open question rather than to infer an answer.
Two patterns carry every observed instance across nine candidates. That is a small sample, and the catalogue is a starting taxonomy rather than a survey of the field.
The four empty entries will stay empty until a real candidate fills them. That is the condition under which the two populated entries can be trusted.