CAVEAT
Controlled Annotation Vocabulary for Epistemic Aberration Types
An OWL 2 ontology for classifying unreliable scientific literature and the markers used to detect it.
@prefix caveat: <https://w3id.org/intellicat/caveat#> .
Download Turtle Download Merged Turtle Download RDF/XML View on GitHub
Purpose
Scientific literature contains unreliable publications ranging from outright fraud to methodological incompetence. CAVEAT provides a shared, machine-readable vocabulary for describing what went wrong and how we know, enabling interoperable tools for research credibility and integrity assessment.
Structure
CAVEAT models two independent dimensions:
- Scientific domain: What field does the paper claim to contribute to? CAVEAT maps the four-level OpenAlex hierarchy (domain, field, subfield, topic) to SKOS concepts, with the sibling lists OpenAlex publishes at each level. The taxonomy ships as two modules the root does not import:
modules/openalexandmodules/openalex-siblings. - Unreliability mode: What kind of failure does the paper exhibit? Modes form a class hierarchy with four top-level categories, each with a primary cause and a default severity.
| Category | Default severity | Covers |
|---|---|---|
| Deliberate misconduct |
1.0
|
Data misconduct (fabrication, falsification, selective reporting), authorship misconduct, publication process misconduct including predatory publishing and compromised peer review, systematic fraudulent production including papermills. |
| Premise-level failure |
0.95
|
Pseudoscience, denialism. |
| Interpretive failure |
0.6
|
Pathological science, systematic misinterpretation. |
| Execution-level failure |
0.3
|
Design failures, statistical malpractice, reproducibility failures, overreach. |
See Default Severity in CAVEAT for details on the severity scale, assignment principles, and consumer overrides.
Detection markers
Detection markers are observable features of publications that serve as evidence for unreliability modes. CAVEAT defines four marker families: textual, data and figure, metadata, and reproducibility. Markers are linked to failure modes through named evidence links, each with its own strength: definitive, strong, moderate, weak, or graded (read from the intensity of each observation). Stated-reason links take their mode from the notice itself, as with retraction notices.
Controlled vocabularies
Controlled vocabularies are domain-specific YAML lexicons that inherit down the mode hierarchy: a child mode inherits its ancestor terms and adds its own. See Vocabulary Inheritance for resolution mechanics, Term Classification for the rejection and sanewashing taxonomy, and Term Frequency Bands for diagnostic frequency levels.
Design principles
- Core classes are aligned to BFO and IAO. Unreliability modes and detection markers are BFO qualities; annotated documents are IAO information content entities. Supporting classes are not aligned.
- FMEA-inspired severity. Every unreliability mode below the root has a default severity between 0.0 and 1.0. Causation is annotated on the four top-level categories.
- Strict TBox/ABox separation. CAVEAT defines the schema, not a corpus.
- Top-down vocabulary inheritance. A child mode inherits its ancestors' terms and adds its own.
- Genus-plus-differentia definitions for every unreliability mode and detection marker.
- Domain classification anchors at the OpenAlex topic level, the most granular of its four levels.
Usage
Downstream ontologies and annotation systems import the merged ontology via owl:imports <https://w3id.org/intellicat/caveat>.
@prefix owl: <http://www.w3.org/2002/07/owl#> .
@prefix caveat: <https://w3id.org/intellicat/caveat#> .
@prefix caveatoa: <https://w3id.org/intellicat/caveat/modules/openalex#> .
@prefix dcterms: <http://purl.org/dc/terms/> .
@prefix ex: <https://example.org/> .
<https://example.org/annotations> a owl:Ontology ;
owl:imports <https://w3id.org/intellicat/caveat> ,
<https://w3id.org/intellicat/caveat/modules/openalex> .
# caveatoa:T13044 is the OpenAlex topic "Biofield Effects and Biophysics".
ex:some_paper a caveat:AnnotatedDocument ;
dcterms:title "Some Dubious Paper" ;
caveat:primaryUnreliabilityMode caveat:BiofieldEnergyHealing ;
caveat:detectionMarkerObserved caveat:PredatoryJournalIndexing ;
caveat:claimedDomain caveatoa:T13044 .
Documentation
Design documents:
- Reporting Profile: Explaining pipeline scores with modes, markers, detectors and confidence
- BFO and IAO Alignment: Upper ontology alignment decisions for modes, markers, and documents
- Default Severity in CAVEAT: FMEA severity scale, assignment principles, and consumer overrides
- Term Classification in CAVEAT Vocabularies: Ordinal ranking from categorical rejection to sanewashing
- Term Frequency Bands in CAVEAT Vocabularies: Domain specificity levels and general-corpus frequency bands
- Vocabulary Inheritance in CAVEAT: Top-down inheritance model and resolution algorithm
Repository specifications and guides:
Download
Ontology distributions:
caveat.ttl: Root ontology, metadata andowl:importsonlycaveat-full.ttl: All modules and imports mergedcaveat.owl: RDF/XML of the merged graph
Ontology modules:
modules/assessment.ttlmodules/openalex.ttl: OpenAlex domains, fields, subfields and topics as SKOS concepts (not imported by the root)modules/openalex-siblings.ttl: the sibling lists OpenAlex publishes (not imported by the root)modules/detection-markers.ttlmodules/document-schema.ttlmodules/marker-evidence.ttlmodules/scientific-domains.ttlmodules/unreliability-modes.ttl
Imports and vocabularies:
Reporting profile caveat-assessment/1:
vocabularies/profiles/assessment-1.jsonld: JSON-LD contextvocabularies/profiles/assessment-1.schema.json: JSON Schema
Citation
CAVEAT is written by the CAVEAT contributors and maintained by Intellicat If you use it, please cite:
CAVEAT contributors. (2026). CAVEAT: Controlled Annotation Vocabulary for Epistemic Aberration Types (Version 0.4.0) [Data set]. https://w3id.org/intellicat/caveat
Machine-readable metadata is in CITATION.cff. The "Cite this repository" button on GitHub generates APA and BibTeX from it.
License
CAVEAT is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).