Glossary

Implemented

Searchable Sita terminology.

Terms

Term	Definition
Sita	A façade-oriented AI decision-intelligence system and research instrument for decarbonisation in AEC. Not documented as a finished production product.
Structural Memory	Project-specific model graphs: elements, geometry, semantic relations and spatial (RCC8) relations derived from IFC/STEP.
Reference Memory	External/contextual knowledge used to reason: materials, cost, lifecycle, codes, and (today) synthetic catalogues. Configured on the Visual Decision Programming canvas.
Decision Memory	The audit trail of a run: invocations, evidence, alternatives, human verdicts, closures and reports.
Planner Agent	Orchestration and synthesis layer. Specialists do not make the project decision; the Planner coordinates them and hands alternatives to a human.
Specialised Agent	A heuristic specialist — Material, Cost, Lifecycle, or Regulation — that reads Reference Memory and reports a finding to the Planner.
Project Knowledge Graph	The intended union of multiple model graphs under one project, with cross-model semantic connections. Today graphs are still keyed per file.
Model Graph	A per-file :Model / :Element graph in Neo4j, produced by structural ingest.
Semantic Relation	Authoring-model relationships such as :CONTAINS and :CONNECTS, distinct from RCC8 spatial predicates.
Geometry Ontology	Sita's façade ontology: element classes, semantic layer, RCC8 spatial layer, reference links and decision semantics.
RCC8	Region Connection Calculus 8. Sita stores JEPD predicates on :SPATIALLY edges. Production classification is AABB unless SITA_GEOMETRY_ENGINE=occt.
Choice Space	The set of candidate alternatives before and after Reference Memory and agent constraints. Visualised as convergence; current charts may use simulated analytical data.
Planner Convergence	The narrowing from a broad candidate space to viable alternatives the Planner can hand to a human.
HITL	Human-in-the-loop. Sita produces informed alternatives; a named person records ACCEPT / REJECT / REQUEST_REVISION.
Break-even	Cost-only lifecycle crossing year: installations(t) = 1 + floor(t / interval); cumulative = element_cost × installations. Carbon is not converted into money.
Graph Analysis	Machine-readable inspection of the project/model graph: counts, coverage, neighbourhoods, filters.
Semantic Analysis	Human-readable views over the graph. Relation bands are implemented; eight other diagrams are planned.
Alternative	A structured option produced by a run — typically a per-element material swap with metrics, evidence and a graph delta.
Evidence	A frozen reference row or citation persisted on a DecisionRun. Not a live query.
Provenance	Five keys: org_id, project_id, model_id, model_version_id, source_element_id. Absence is a statement, not a missing column.
Visual Decision Programming	The Reference Memory canvas: agents, sources and skills wired as an acyclic evidence graph. This is the research instrument's configurable AI architecture.
Epistemic label	How a value is known: deterministic_geometry, rule_derivation, synthetic_data, real_evidence, model_inference, llm_reasoning, human_choice.
10-step framework	Canonical data-flow steps 1–9 in docs/sita-master-architecture §5. Step 10 is not named there.