| Agent |
AI-enabled component that plans or performs tasks, often using models, memory, retrieval, or tools. |
| Agent ID/version |
Stable identity and version of the agent configuration that performed work. |
| Applicability |
Whether the subject of a requirement exists within the approved system scope. The playbook's “Not applicable” record status is distinct from an ORF matrix action or waiver. |
| Audit trail |
Ordered records used to establish who/what acted, what changed, and what outcome occurred. |
| Category code |
Short ORF requirement prefix identifying an operational category: AGT, PERF, UX, SYS, SEC, or GOV. For example, AGT represents the agent-level observability category in IDs such as ORF-AGT-04. |
| Correlation |
Connecting related logs, spans, metrics, cases, changes, and evidence through stable IDs. |
| Evidence coverage |
Proportion of expected events or journeys for which required evidence is present. |
| Good-to-Have |
Encouraged item not required by the ORF matrix for the system risk tier. |
| Groundedness |
Degree to which a response's material claims are supported by the intended evidence sources. |
| Hallucination |
Materially incorrect or unsupported output under an approved domain definition and review protocol. |
| Intermediate step |
Observable subtask, tool plan/action, decision, or system-produced summary; not hidden model chain-of-thought. |
| Mandatory |
Requirement that must be implemented with no waiver permitted. |
| Mandatory but Waivable |
Requirement that must be implemented unless authorised, documented risk acceptance is approved. |
| Metric contract |
Definition of formula, source events, window, dimensions, coverage, version, threshold, owner, and response. |
| Query |
User-issued request that initiates work. |
| Request |
A specific model, retrieval, tool, or service operation within a task. |
| Residual risk |
Risk remaining after controls and compensating measures are applied. |
| RAG |
Retrieval-augmented generation, where retrieved information is supplied as context for generation. |
| Safe failure |
Task is not completed but the system declines, hands over, or stops without creating an unsafe outcome. |
| Session |
Complete interaction context containing one or more queries and downstream tasks/requests. |
| Span |
Timed operation within a distributed trace, linked by trace, span, and parent relationships. |
| SRE |
Site reliability engineering. |
| SLO |
Service-level objective. |
| Task |
Goal attempted in response to a query, possibly containing subtasks and multiple requests. |
| Telemetry priority tier |
ORF's intrinsic measure of how foundational an item is; separate from system risk tier. |
| Trace |
Connected execution record across components and operations. |
| Unsafe success |
Task is completed but produces a policy-violating, harmful, or otherwise unsafe outcome. |
| Waiver |
Time-bound authorised decision to accept residual risk for a waivable requirement. |