Multi-Agent System Ontology

Abstract base for all analysis-layer computed artefacts. dataProvenance = 'computed'. Not emitted during execution — produced by offline or streaming analysis pipelines.

Inlier/outlier classification summary for one analysis layer (text/graph/metric), computed across all sessions in a SemanticGroup. Session/MetricResult membership (outlier or not) is via mas:isAnomalous edges; mas:scores/mas:threshold hold the supporting numeric detail.

How one Cognition protocol went in one Session: which agents had it and which used it, and the protocol's evaluation metrics, as MetricResults that measure this report. One per Session and protocol.

Session-level summary of the Cognition protocols: which were enabled (available to the session's agents) and which were activated (actually used). Lists protocol names, so a new protocol adds an entry rather than new properties. One per Session.

A cognitive failure (from the challenges taxonomy, e.g. IR-1 or CC-3) detected in a session: the challenge code, the detector's confidence, and a brief reasoning. Attached to the Session, and to the call where the challenge happened when it applies to one.

Statistical consistency summary for one analysis layer (text/graph/metric) across all sessions in a SemanticGroup: a mean, range, and confidence interval/indicator over the per-session consistency statistic.

Representative/centroid summary of normal behaviour for one analysis layer, computed across all sessions in a SemanticGroup.

Common parent for analysis reports computed over an entire SemanticGroup (as opposed to a single Session or MetricResult). A report cannot exist without the group it was computed over.

The Metric this report was computed over. Only present when mas:dataType = 'metric' — textual-output and execution-graph reports leave this unset rather than carrying a metric-specific field they don't need. See mas:AnalysisElementAboutMetricShape below, which enforces that conditional requirement.
Domain: AnalysisElement
Range: Metric

Names of the agents that took part in the protocol: they actually used it.
Domain: CognitionProtocolExecutionReport
Range: string

Names of the agents that had the protocol available.
Domain: CognitionProtocolExecutionReport
Range: string

The call (LLM, tool or agent call) where the challenge happened. Absent for challenges that only show at the level of the whole session.
Domain: CognitiveFailureReport
Range: ExecutionElement

Centroid of the group's sessions for this layer — a scalar value, or JSON-encoded if structured.
Domain: NormalBehaviourReport
Range: string

Code of the detected challenge in the challenges taxonomy, e.g. 'IR-1', 'DC-4', 'CC-3'.
Domain: CognitiveFailureReport
Range: string

How sure the detector is that the challenge happened: 'low' | 'medium' | 'high'.
Domain: CognitiveFailureReport
Range: string

Human-readable confidence label for this report, e.g. 'low' | 'medium' | 'high'.
Domain: ConsistencyReport
Range: string

JSON-encoded confidence interval dict (e.g. {'lower': .., 'upper': ..}) around the mean.
Domain: ConsistencyReport
Range: string

What kind of data this report was computed over: 'text' | 'graph' | 'metric'. Discriminates whether mas:aboutMetric is populated.
Domain: AnalysisElement
Range: string

How we know the protocol was enabled: 'system_prompt' when an agent's prompt lists one of its skills, 'activation' when only a call to it shows it was there.
Domain: CognitionProtocolExecutionReport
Range: string

Links an outlier Session, or an outlier MetricResult to the AnomalyReport that flagged it.
Domain: n88ae075d7d9c4bfc9ce14e6ac9be439bb7
Range: AnomalyReport

Mean value of the consistency statistic across the group's sessions.
Domain: ConsistencyReport
Range: decimal

Number of MetricResults currently measuring this report, from the protocol's latest evaluation.
Domain: CognitionProtocolExecutionReport
Range: integer

The SemanticGroup this report was computed over. A SemanticAnalysisReport cannot exist without one; a SemanticGroup does not require any reports.
Domain: SemanticAnalysisReport
Range: SemanticGroup

Name of the Cognition protocol this report is about, for example 'accord'.
Domain: CognitionProtocolExecutionReport
Range: string

Names of the Cognition protocols that were activated: an agent actually used them (for example, called one of their skills).
Domain: CognitionProtocolReport
Range: string

Names of the Cognition protocols that were enabled: available to at least one agent of the session (for example, listed among its skills).
Domain: CognitionProtocolReport
Range: string

JSON-encoded raw result dict produced by the normal-behaviour detector (layer-specific shape).
Domain: NormalBehaviourReport
Range: string

When this report was last written.
Domain: n88ae075d7d9c4bfc9ce14e6ac9be439bb14
Range: dateTime

The Session this session-level report (Cognition protocol or cognitive failure) describes. A report cannot exist without one.
Domain: n88ae075d7d9c4bfc9ce14e6ac9be439bb10
Range: Session

Processed/normalised form of the representative sample, if the detector produced one.
Domain: NormalBehaviourReport
Range: string

Identifier (or JSON-encoded value) of the session/sample closest to the centroid — the group's representative example.
Domain: NormalBehaviourReport
Range: string

JSON-encoded list of raw per-session anomaly scores backing the inlier/outlier classification.
Domain: AnomalyReport
Range: string

Names of the protocol's skills that agents actually called.
Domain: CognitionProtocolExecutionReport
Range: string

Names of the protocol's skills that were available to the session's agents.
Domain: CognitionProtocolExecutionReport
Range: string

Score threshold above which a session was classified as an outlier.
Domain: AnomalyReport
Range: decimal