Module bundle 

Content-addressed finding bundles: the atomic object of the Vela protocol.

Structs

Annotation

A lightweight annotation on a finding — like a comment on a line of code.

Artifact

Generic content-addressed artifact.

Assertion

Attachment

A file attached to a finding (dataset, figure, supplementary material).

Author

CodeArtifact

v0.33: CodeArtifact as a first-class kernel object.

Conditions

Confidence

ConfidenceComponents

Structured breakdown of frontier epistemic confidence (v0.2.0).

ConfidenceUpdate

A confidence update is a mutable interpretation layer event. The finding’s evidence is immutable; the confidence assessment can evolve.

Dataset

v0.33: Dataset as a first-class kernel object.

Entity

Evidence

Extraction

FindingBundle

Flags

Link

NegativeResult

Prediction

v0.34: Prediction as a first-class kernel object.

Provenance

ProvenanceRef

Phase β (v0.6): structured provenance reference attached to an annotation (or any future note-shaped object). At least one identifying field (doi, pmid, title) must be set when the provenance is present; an all-empty ProvenanceRef is rejected by validate_event_payload.

Replication

v0.32: Replication as a first-class kernel object.

Resolution

v0.34: Resolution closes out a Prediction.

ResolvedId

A resolved identifier from a scientific database.

Review

ReviewEvent

A review event is a content-addressed record of human judgment on a finding. Like a Git commit, it records who, when, what changed, and why.

ReviewEvidence

Trajectory

TrajectoryStep

Enums

CausalClaim

v0.38: Pearlian causal typing for an assertion. The kernel’s pre-v0.38 record carried only direction: Some("positive" | "negative") — enough to know that “X covaries with Y” but not whether the speaker meant correlation, mediation, or intervention. In real review work those are different epistemic claims with different evidence requirements; conflating them produced silent over-claiming.

CausalEvidenceGrade

v0.38: study-design grade backing a causal claim. The grade is what makes the difference between “the data is consistent with X causing Y” (Observational) and “X causes Y” (Rct). The kernel carries the design label so reviewers can re-grade without re-extracting.

ConfidenceKind

Semantic category of the confidence score stored on the frontier.

ConfidenceMethod

Confidence method: how the score was determined.

ExpectedOutcome

v0.34: ExpectedOutcome — the structured shape of a Prediction’s expected resolution.

LinkParseError

LinkRef

v0.8: typed reference resolved from Link.target.

Mechanism

v0.45: structural causal mechanism on a directed edge.

MechanismSign

v0.45: causal direction on a Mechanism.

NegativeResultKind

v0.49: NegativeResult as a first-class kernel object.

ResolutionMethod

How an entity was resolved to its canonical form (v0.2.0 schema).

ReviewAction

The action taken in a review event.

ReviewState

Typed review state. Replaces the v0.2 flags.contested: bool collapse of three semantically distinct review judgments. Doctrine line 6: “scientific disagreement should remain live state.”

TrajectoryStepKind

v0.50: Trajectory as a first-class kernel object.

Constants

VALID_ARTIFACT_KINDS

Valid artifact kinds for the generic Artifact kernel object.

VALID_ASSERTION_TYPES

Valid assertion types per schema.

VALID_CAUSAL_CLAIMS

Valid string forms for serialized CausalClaim. The kernel validates against this on load.

VALID_CAUSAL_EVIDENCE_GRADES

Valid string forms for serialized CausalEvidenceGrade.

VALID_ENTITY_TYPES

Valid entity types per schema. Single source of truth shared by the validator and the vela finding add CLI; do not duplicate.

VALID_EVIDENCE_TYPES

Valid evidence types per schema.

VALID_LINK_TYPES

Valid link types per protocol §5.

VALID_PROVENANCE_SOURCE_TYPES

Valid provenance source types per schema.

VALID_REPLICATION_OUTCOMES

Valid replication outcomes per v0.32 schema.

Functions

causal_consistency_multiplier

v0.38.1: causal-claim × evidence-grade compatibility multiplier. An RCT supports any claim slightly better than baseline; an observational study weakly supports correlation; an intervention claim from observational data gets a meaningful penalty (the design doesn’t actually identify the causal effect being claimed). Returns 1.0 when either field is None — the pre-v0.38 case is neutral.

compute_confidence

Compute frontier epistemic confidence from evidence and condition fields. Returns a fully populated Confidence with components and aggregate score, using a deterministic, auditable support computation.

compute_confidence_from_components

Pure-math kernel for the frontier-epistemic confidence formula. Takes replication counts and (v0.38.1) the optional causal typing as inputs so the same math drives both the legacy scalar path (compute_confidence) and the v0.32+ Project-aware path (Project::compute_confidence_for).

count_replication_outcomes

Count v0.32 replication outcomes targeting a given finding id. Returns (n_replicated, n_failed, n_partial). Inconclusive outcomes are deliberately excluded — they represent methodological ambiguity and don’t move the confidence score.

recompute_all_confidence

Recompute confidence scores for all findings in a slice using the v0.32 Replication collection as the source of truth. Returns the number of findings whose score changed by more than 0.001.

valid_artifact_kind