Search for a command to run...
This document serves as the companion reproducibility package for the manuscript Active Learning (AL)-Related Mechanisms in LLM-Based Code Generation: A Systematic Literature Review and Design Framework. Its purpose is to document, in a transparent and auditable form, the full search, screening, and selection procedure that produced the final evidence base used in the review. To support reproducibility, the document consolidates the database search strategy, PRISMA audit trail, and layer-wise evidence distribution across the three review layers: (i) AL Foundations and AL-relevant general machine learning, (ii) AL-related mechanisms in LLM-based code generation, and (iii) AL-relevant evidence for PLC-oriented code generation and safety-critical control synthesis. The formal search was restricted to two peer-reviewed databases, Scopus and IEEE Xplore. In both databases, searches were limited to metadata fields only, namely Title, Abstract, and Keywords, to ensure a consistent and defensible basis for retrieval. The canonical Boolean structure was kept constant across databases, while the execution syntax was adapted to each platform: TITLE-ABS-KEY(...) in Scopus Advanced Search and "All Metadata": commands in IEEE Xplore Advanced Search. The documented search returned 711 raw records in total, comprising 386 records from Scopus and 325 records from IEEE Xplore. At layer level, the raw search yielded 300 records for Layer~(i), 243 for Layer~(ii), and 168 for Layer~(iii). After deduplication, 76 duplicate records were removed, leaving 635 records for title--abstract screening. Of these, 280 records were excluded at the initial screening stage, and 355 reports were taken forward for retrieval. After9 reports could not be obtained, 346 full texts were assessed for eligibility. A further 194 studies were excluded after full-text review, resulting in a final qualitative corpus of 152 studies. The final evidence base is distributed across the three review layers as follows: Layer~(i) AL Foundations and AL-relevant general machine learning (n = 60), Layer~(ii) AL-related mechanisms in LLM-based code generation (n = 52), and Layer~(iii) AL-relevant evidence for PLC-oriented code generation and safety-critical control synthesis (n = 40). Overall, this document is intended to function as a standalone replication record. It enables an external researcher to reconstruct the search logic, verify the PRISMA counts, trace the database-level contributions from Scopus and IEEE Xplore, and reproduce the final evidence base used for the review's synthesis and framework derivation.