Fault data scarcity in HVAC fault detection: systematic review data and code

Data, protocols, checking files and the grading script behind the systematic review Fault data scarcity in HVAC fault detection: a critical review of transfer learning, synthetic data and foundation models by R. Simson, K.-V. Võsa, M. Kiil and J. Kurnitski, together with the supplementary material of the review. The review covers 209 primary studies and 11 reviews on three ways of training fault detection and diagnosis models for building heating, ventilation and air-conditioning (HVAC) systems when labelled fault data are scarce: transfer and few-label learning, synthetic fault data, and foundation models. It grades eight main claims by how many independent research groups support them, with public data and with tests on data separate from training. Files Supplementary_material.pdf and Supplementary_data.xlsx: the supplementary material of the review (text supplements S1 to S3, S8, S11 and S13 to S15; tables S4 to S7, S9 and S10). data.zip: the extraction table of every record assessed at full text (S4, CSV), every screened record and its screening decision (S5, S6), the PRISMA 2020 counts (S7, JSON), the evidence tables (S9, S10), the full references of the included studies (S14) and the evidence grades of every claim (evidence_grades.json). protocols.zip: the index of the supplementary material (S0), the search queries (S1), the screening and extraction protocols (S2, S3), the extraction checks (S8), the full methodology (S11), the PRISMA 2020 checklist (S13) and the evidence analyses (S15), as Markdown. checks.zip (supplement S12): the check of the extracted data against the full texts, one row per study and field, and the records found by the Semantic Scholar search. code.zip: grade_claims.py, which rebuilds the evidence grades from S4. Run python code/grade_claims.py from the unzipped deposit; the output is identical to data/evidence_grades.json. Not included No full text of any study; each record carries its DOI. Abstracts only where Crossref or OpenAlex serves them openly (S5, column abstract_source). The search, screening-support and full-text retrieval scripts, which need licensed API access and local copies of the full texts. The search strings are given verbatim in S1. Licence: the authors' compilation, codes, labels and text under CC BY 4.0; the code (code/grade_claims.py) under the MIT licence. See README.md and LICENSE.md.