Abstract
<p>Buildings dominate final energy demand, but operation remains reactive. This paper reviews cognitive and proactive building operation and derives a research agenda. Two questions guide the analysis: RQ1, the current status of seven constituent streams; RQ2, the gaps explaining the persistent distance between concept and practice. A critical review (July 2026) covers cognitive building concepts, MPC, RL/data-driven control, OCC, AFDD, cognitive digital twins with semantic infrastructure, and LLM-based agentic systems. Retrieval prioritised recent reviews and field deployments, with snowballing. An explicit evidence hierarchy and gap-derivation rule are stated, and bibliographic records were verified against metadata. An independent bibliometric strand uses a documented Web of Science query on cognitive, adaptive, cybernetic, and AI vocabulary, yielding 113 records (2011-2026). This corpus is profiled bibliometrically and mapped onto the gap taxonomy. The query string is reproduced verbatim, its retrieval behaviour is audited, and two artefacts materially affecting interpretation are discussed. The status pattern is consistent: conceptually rich, simulation-mature, field-poor. Bibliometrics sharpen this: 49% of records are from 2024-2026; a single 2020 review accounts for 49% of citations; only 20% are building-related (rest in software, manufacturing, general digital twins); cybernetic-building intersections contain only three records, the latest from 2015; co-word analysis reveals structural zeros, with cybernetics and self-adaptive systems never co-occurring. The query audit confirms: although the viable system model is named, exactly one of 113 records engages Beer's model, and it does not concern buildings. Outside the database, this organizing principle is applied in production management, software architecture, and agentic AI, but not in contemporary building operation. Eight gaps are derived: G1 definitional/theoretical, G2 reactive-to-proactive practice, G3 simulation-to-reality, G4 fragmentation, G5 architectural mechanism, G6 evaluation, G7 longevity/resilience, and G8 human factors/governance. G1-G4 and G6-G8 restate explicit deficits; G5 is an interpretive synthesis. The paper argues G5, the absence of an organizing principle for allocating sensing, computation, and actuation across subsystems, spatial scales, and time horizons, is structurally prior to several other gaps, and candidate principles are compared. Contributions include a cross-stream status consolidation, an evidence-anchored gap taxonomy with derivation rules, bibliometric quantitative support, and a defended keystone-gap argument. The review is evaluative rather than exhaustive and positions its findings as an evidence-based hypothesis set for a subsequent protocol-driven systematic review.</p>