The Sovel vocabulary
Precise definitions for the concepts, metrics, and structures that appear throughout the Sovel product and documentation.
Risk types surfaced by the gap engine from work order history
Knowledge concentration
A condition where one or two technicians account for the majority of successful resolutions on a critical asset. Measured by assignee dominance ratio and resolution-time variance when the expert is absent. High concentration means a single retirement or departure creates an immediate operational gap.
Retirement risk
An issue subtype scored by combining expert tenure with concentration score and asset criticality. Long-tenured experts with deep, undocumented knowledge on high-criticality assets represent the highest-urgency capture targets. Retirement risk issues are prioritized at the top of the issue board.
Procedure drift
A measurable divergence between what technicians write in work order notes and what the documented procedure says to do. Drift is a governance signal — not automatically a problem. The reviewer determines whether to update the SOP (if field practice is better) or retrain to the existing procedure (if the drift is a quality risk). The phrase "the field story and the book diverge" captures the dynamic.
Shadow work
Undocumented workarounds that appear in WO narrative text but are not reflected in any official procedure or knowledge entry. Shadow work often represents the real institutional knowledge — the informal fix that works when the manual approach fails. Capturing and reviewing it either formalizes a valid workaround or flags a safety deviation for correction.
Recurring failures
Work orders for the same asset and failure mode that repeat without a documented root cause or resolution procedure being established. Each recurrence is evidence that institutional knowledge exists (someone fixed it) but was not captured. High recurrence without structured closure is one of the strongest signals of undocumented expertise.
The human-decision layer that governs what becomes operational truth
Reviewer primacy
The foundational design principle of Sovel: no knowledge object advances from draft to governed state without an explicit human approval. AI models surface candidates, structure drafts, and summarize evidence. Sign-off, safety culture, and the change process belong to the reviewer. Reviewer primacy is not a guardrail added on top — it is the architecture.
Point-of-decision capture
Capturing expert knowledge at the moment it is most accessible — when a technician is completing a work order on an asset they know well. At that moment, the context is fresh, the motivation is concrete, and the contribution is tied to a specific event the system can verify. Point-of-decision capture is more reliable than retrospective interviews.
Knowledge-governance-first
The product philosophy that prioritizes governance of knowledge over volume of knowledge. A large repository of ungoverned, unverified content is a liability, not an asset. Sovel deliberately constrains what can become an Operations Skill — only reviewed, approved, and auditably attributed knowledge passes through. Quality over quantity, always.
Propagation alert
A system notification raised when a proposed knowledge entry contradicts or materially conflicts with an existing governed Operations Skill. Propagation alerts prevent silent overwriting of institutional memory. The reviewer resolves the conflict — the AI does not auto-arbitrate. Conflict resolution decisions are logged with reason codes.
Reason code
A structured, mandatory annotation attached to every reviewer decision (APPROVE, EDIT, REJECT, DEFER). Reason codes preserve the "why" behind governance decisions. They enable audit reconstruction — an engineer months later can understand not just what was decided, but the context and rationale. They also inform correction inference over time.
Metrics that track whether the organization is getting smarter
Knowledge metabolism
A measure of whether knowledge moves end-to-end through the detect → capture → govern → place loop. High metabolism means detected risks are flowing through to governed placements efficiently. Low metabolism signals a bottleneck — usually at capture (not enough expert contributions) or governance (reviewer throughput is too low).
Net knowledge position
A scalar score representing whether the organization's governed knowledge base is growing (positive) or decaying (negative). Computed from the ratio of new placements to stale entries, weighted by asset criticality. A positive net knowledge position means the organization is getting smarter faster than it is forgetting.
Expert concentration
The degree to which a given asset or failure mode is exclusively understood by one person. Distinct from "knowledge concentration" (which measures assignee patterns in WOs) in that expert concentration is measured against the governed knowledge graph — specifically, how many unique approved experts are linked to each asset node. A single-expert asset is maximum concentration risk.
Knowledge tenure
The duration for which an organization retains usable, governed operational expertise — independent of individual headcount. The Sovel mission is to extend knowledge tenure: to ensure that what experienced maintainers know outlasts their employment. This is what "Sovel" refers to — the Latin root relating to tenacity, persistence, and holding on.
Data objects and system components in the Sovel architecture
Operations Skill
The atomic, versioned unit of governed operational knowledge in Sovel. An Operations Skill binds a specific failure mode and asset context to structured resolution guidance, along with full attribution (author, reviewer, reason code, timestamp) and a confidence level. Skills are placed into the Maintenance Ontology and are the basis for knowledge graph queries, staleness tracking, and MTTR outcome linkage.
Author: R. Delgado · Reviewer: J. Miller · v3 · Confidence: HIGH
Maintenance Ontology
A typed graph of assets, failure modes, experts, Operations Skills, and the relationships between them. The Maintenance Ontology is the placement target — when a reviewer approves a skill, it is bound to the relevant node in the ontology. The graph enables relationship queries ("what expertise breaks if this person retires?") that flat databases cannot answer.
Correction inference
The process by which Sovel learns from reviewer edit patterns to improve future AI-structured drafts. When a reviewer consistently modifies the same field or dimension across multiple entries, the system infers that the extraction model needs adjustment in that dimension. Correction inference improves draft quality without overriding reviewer authority — it is a feedback mechanism, not an autonomous learning override.
Lint loop
A periodic AI-agent scan of the governed knowledge base that checks for staleness (entries not revalidated within their confidence window), contradictions (skills that conflict with each other or with newer WO evidence), and coverage gaps (assets with high failure frequency and no governed skill). Lint findings are proposed back to the reviewer inbox — not auto-corrected. The reviewer decides; the lint loop surfaces candidates.