Supplementary Data for Small Language Models for AEC Tasks: Technology Evolution, Current Practices, and Future AI Systems

Abstract

# Supplementary Data Package This package contains this README and four supplementary workbooks needed to trace the study corpus, screening validation, reporting, and the evidence summarized in manuscript Tables 5 and 6. `Supplementary_Table_S1_Corpus_and_Search.xlsx` documents the evidence base used in the review. Its `Corpus_132` worksheet lists the 132 included studies with bibliographic identifiers and task classifications, while `Seed_Reviews_22` records the 22 review-derived seed sources and their role in retrieval. The workbook supports the corpus construction, study identification, and task-classification information described in Section 3.1 and used throughout the evidence map. The seed set documents one retrieval route and is not treated as an independent recall-estimation sample. `Supplementary_Table_S2_Screening_Validation.xlsx` documents the screening-validation exercise. It reports agreement statistics for a 40-record sample, preserves the archived AI decisions and the scores and inclusion decisions of two independent reviewers, and defines the screening categories. The workbook supports the screening-validation procedure and agreement results reported in Section 3.1. The original blinded decisions remain separate from the later adjudication results. `Supplementary_Table_S3_Reporting_and_Risk_Audit_132.xlsx` provides the study-level reporting for all 132 included studies. It covers parameter disclosure, operational SLM eligibility, inference and deployment reporting, public availability of evaluation materials, baseline comparability, test-set size, run repetition, prompt disclosure, model or API version disclosure, and decoding settings. Definitions and page-labelled supporting evidence are retained in the same workbook. These data support the reporting that qualify the comparative synthesis, including the study counts reported at the beginning of Section 3.4 and the statements concerning public data, comparable baselines, test-set reporting, repeated runs, prompts, and model versions. `Supplementary_Table_S4_Table5_Table6_Evidence.xlsx` contains the evidence used for the two principal SLM syntheses. Its `Table5_Evidence` worksheet provides 29 task-level comparison records from 23 studies, including task class, evaluation standard, A0/A1/A2 adaptation profiles, P1/P2 models, selected results, and bounded interpretations. Its `Table6_Eligibility` worksheet lists 32 formal empirical SLM studies: 29 with a verified focal generative model below 10B and three admitted through an official mini designation whose parameter count was not disclosed. The first worksheet supports manuscript Table 5 and Section 3.4, whereas the second supports manuscript Table 6 and Section 3.5. Table 5 is a comparative task-level subset; Table 6 is a broader study-level inventory and does not require a larger-model comparator. The identifiers `AEC-001` to `AEC-132` are stable across the workbooks, and citation keys and DOIs link the records to the manuscript bibliography. `NA`, `unknown`, and related missingness codes are retained when a paper does not report a value or when a comparison is not applicable; unreported information is not imputed. The definitions included in each workbook govern the interpretation of its coded fields. The workbooks provide supporting evidence rather than replacing the manuscript tables: Tables 5 and 6 present condensed syntheses, while Supplementary Table S4 preserves task- and study-level traceability. Detailed screening and coding described in the manuscript supplementary appendices are not duplicated in this data package.