Glossary

Candidate — an element the AI has extracted or proposed that is not yet part of the model. Candidates are reviewed by risk tier before they count.

Closure chain — the traced line behind a resolved gap: gap → recommendation → target process → data objects → application → capability. A critical gap can’t be marked resolved until its chain is complete.

Commitment tier (Tier 1 / 2 / 3) — how a candidate is routed for review. Tier 1 (high confidence, multi-source, low-impact) auto-commits; Tier 2 gets batch human review; Tier 3 gets individual review. Gaps, recommendations, and objectives are always Tier 3.

Deliverable — a consulting artifact eavantage generates from the model — a report, map, spec, or deck. See Deliverables.

Domain pack — a bundle of content (standards, templates, a capability set, a role) for a particular kind of engagement. See Domain packs. Not to be confused with the metamodel — a pack changes what you start with; the metamodel changes what types and views exist at all.

Engagement — one piece of work for one client, moving through the six phases.

Gate — a server-side check between phases. A phase doesn’t advance until its gate passes.

Governed loop — the core cycle: material in → AI proposes candidates → candidates reviewed → approved elements become authoritative. See The governed loop.

Metamodel — the rulebook the model of record is built on: which entity types and relationships exist, their scoring dimensions, and which viewpoints are available. Configurable deployment-wide and per engagement, without weakening the single-writer rule, tenant isolation, or the approval gate. See The metamodel.

Model of record — the authoritative model of the client’s architecture. Only approved elements enter it, and only one software component writes it.

Provenance link — the link a candidate carries back to the exact passage in the source document it came from.

Readiness score — the Phase 2 output summarising how ready the organisation is against the assessed dimensions.

Seal — the state an engagement enters at closure: deliverables and templates locked, audit trail read-only.

Single-writer rule — exactly one software component may write to the authoritative graph. Nothing else can, on any code path.

Standards library — the firm’s curated body of frameworks, patterns, and playbooks, in three tiers. See The standards library.

Tenant — the isolation boundary. One tenant = one end client. Different clients are fully partitioned; multiple engagements for the same client share that client’s graph.

Viewpoint — a view over the model — a matrix, catalog, decomposition tree, network map, flow, or timeline — bound to any slice of the metamodel, including custom types. Renders read-only from the current, approved model.

Vision pipeline — the multi-stage process that turns a diagram file (Visio, draw.io, BPMN, a diagram in a slide) into candidate model elements.