Evidence Vocabulary v1.0-draft

Draft Status: This vocabulary is pre-stable (per D-PATH). The namespace URI is the stable major (evidence/v1#) but the version-info string is 1.0-draft. Pre-stable drafts are NOT registered in spec/VOCAB_VERSIONS; they land there only at v1.0 graduation. The schema may change before that point.

Per-term status: every term in this vocabulary carries vs:term_status (W3C SemWeb Vocabulary Status, originating in FOAF). Draft terms are unstable and deprecated terms are archaic. The status is per TERM by design, so an individual term can be promoted to testing or stable in place, without a namespace change and without graduating the whole vocabulary. Read the status of the term you intend to use rather than the status of the file it lives in.

Submit feedback

Overview

Namespace URI:: https://ns.cascadeprotocol.org/evidence/v1#
Preferred Prefix:: evidence:
Version:: 1.0-draft.0.3 (2026-08-20)
Status:: Draft
Layer:: Layer 2 (cross-cutting domain vocabulary)
Imports / reuse:: cascade: (Core v1 provenance primitives), W3C PROV-O
Schema files:: evidence.ttl (OWL ontology) · evidence.shapes.ttl (SHACL shapes)

The Evidence Vocabulary is a Layer 2 cross-cutting vocabulary for grounding assertions in cited evidence with an explicit outcome. It generalizes the assertion, evidence, verdict pattern needed by Cascade Workbench (checking claims pulled from AI conversations against a patient's records and literature), the Cascade Appeal system (citing clinical evidence against policy criteria), and PGx/genomics decision support.

It reuses core provenance primitives (cascade:extractionModel, cascade:extractionConfidence, cascade:requiresUserReview, prov:wasDerivedFrom) rather than redefining them. The term Assertion is used (not "Claim") to avoid collision with coverage:ClaimRecord.

Download TTL Download Shapes All vocabularies

What this vocabulary covers

Assertion ≠ Evidence ≠ Grounding. An Assertion is a discrete, checkable statement extracted from a source (e.g. an AI conversation) or entered by the user. An EvidenceLink is a reified link from the assertion to one piece of evidence carrying a stance. A GroundingActivity is the PROV-O Activity (the model pass) that graded the assertion.
Reified evidence links. Each EvidenceLink is a node so a single Assertion can carry multiple, differently-directed pieces of evidence. Links point at a Pod record (citesRecord, open range for cross-namespace linking) or an external Citation (citesSource).
External citations as first-class Pod resources. A Citation (identified by DOI and/or PMID) can be retrieved into the Pod (recommended path literature/) so an evidence chip resolves to it exactly as it resolves to a clinical record.
Anti-confirmation-bias by design. disconfirming captures "what would argue against this" considerations, required by product design for any material assertion. rationale carries a one-line plain-language explanation.

The facet model (verdict taxonomy v2, v1-draft.0.2)

The grounding outcome moved from a single flat evidence:verdict to orthogonal facets on the Assertion, each an object property over a closed enumeration, plus a numeric confidence. The facets are the canonical serialized form; the flat evidence:verdict and its VerdictValue individuals are DEPRECATED and kept for one release (see below).

Facet property	Range (closed enum)	Individuals
`evidence:direction`	`DirectionValue`	`Supports`, `Contradicts`, `Mixed`, `None` (what the evidence, in aggregate, says).
`evidence:basis`	`BasisValue`	`Record`, `Literature`, `RecordAndLiterature`, `None` (where the evidence is).
`evidence:strength`	`StrengthValue`	`Strong`, `Moderate`, `Weak`. On basis `Record` this is a deterministic function of the match (guardrail G-2), never model-emitted.
`evidence:settled`	`SettledValue`	`Settled`, `NeedsEvidence` (whether the assertion is resolved).
`evidence:reason`	`NeedsEvidenceReasonValue`	`NoRecord`, `NeedsLiterature`, `NotCheckableByNature` (why it still needs evidence).
`evidence:confidence`	`xsd:decimal` [0.0, 1.0]	Confidence in the grounding result. Distinct from `evidence:groundingConfidence` (the activity's raw model confidence).

Shared individuals. evidence:Supports and evidence:Contradicts are typed both StanceValue (per link) and DirectionValue (aggregate); evidence:None is both a DirectionValue and a BasisValue; evidence:NeedsLiterature is both a deprecated VerdictValue and a NeedsEvidenceReasonValue. The property an individual appears on disambiguates its role.

Deprecated: the flat verdict (removal at v1.0)

evidence:verdict (an object property) and the evidence:VerdictValue individuals (Supported, Contradicted, Unverifiable, NeedsLiterature) are DEPRECATED in v1-draft.0.2, not removed. Draft-period data and fixtures already carry evidence:verdict, and the SHACL shapes keep a legacy branch so that data still validates during the one-release transition. Clients derive the flat value from the facets in code and MUST NOT serialize it going forward. Both are scheduled for removal at v1.0 graduation.

Key classes

Class	Anchor	Purpose
`evidence:Assertion`	subClassOf `prov:Entity`	A discrete, checkable statement evaluated against evidence. Carries the facets and zero-or-more evidence links.
`evidence:EvidenceLink`	OWL Class	Reified link from an Assertion to one piece of evidence (Pod record or Citation) with a stance.
`evidence:Citation`	subClassOf `prov:Entity`	External bibliographic source (DOI/PMID), a first-class Pod resource once retrieved.
`evidence:GroundingActivity`	subClassOf `prov:Activity`	The model pass that graded an Assertion. Reuses `cascade:extractionModel` to record the model.
`evidence:DirectionValue` / `BasisValue` / `StrengthValue` / `SettledValue` / `NeedsEvidenceReasonValue`	OWL Class	The five facet enumerations introduced in v1-draft.0.2.
`evidence:StanceValue`	OWL Class	Per-link direction: `Supports`, `Contradicts`, `Contextual`.
`evidence:VerdictValue`	OWL Class (deprecated)	The flat verdict enumeration, superseded by the facet model.

Validation (SHACL)

evidence.shapes.ttl is expressed in SHACL Core (sh:or / sh:not / sh:in), never SHACL-SPARQL, so that rdf-validate-shacl (and cascade validate) actually enforces every constraint. Three node shapes: AssertionShape, EvidenceLinkShape, CitationShape.

Generalized grounding invariant. A grounded result (settled Settled, direction one of Supports/Contradicts/Mixed, basis not None) of EITHER basis requires at least one evidence:hasEvidenceLink. This generalizes the v0.1 invariant (which only fired for the flat Supported/Contradicted verdicts).
Facet consistency. A NeedsEvidence assertion must not carry a grounded direction (direction must be None); a grounded direction must carry a real basis (basis must not be None).
Graded-ness. Every assertion carries the facet form (settled) or, for draft-period data, the deprecated flat verdict. The legacy branch is retained one release and removed at v1.0.
Bounded confidence. evidence:confidence is an xsd:decimal constrained to [0.0, 1.0].

References

Per-vocab changelog: spec/ontologies/evidence/CHANGELOG.md
Consumed by Workbench v1-draft (the Claims Ledger grounding model) and the Cascade Appeal system.
Schema reference: cascade-protocol-schemas.md (Evidence section).

Provide Feedback

This vocabulary is in draft status. We welcome feedback on the facet model, the grounding invariant, and any gaps for your grounding or decision-support use case. Submit Feedback

Evidence Vocabulary v1.0 Draft (v1.0-draft.0.3)