Published life cycle assessment (LCA) results for building facades and enclosures are difficult to compare because of differences in system scope, methods, and reporting. This systematic review synthesizes cradle-to-gate embodied carbon results by system type and assesses their comparability. Following PRISMA 2020, searches of five databases, supplementary searches, and citation tracking (2010 to January 2026) identified 146 studies with 800 options in 13 system types. The primary dataset contained 597 options with strict A1–A3 results per square meter of facade or enclosure. Their median global warming potential (GWP) ranged from 9.1 kg CO₂e/m² for insulation components to 207 kg CO₂e/m² for building-integrated photovoltaic (BIPV) modules. Opaque walls, the largest category (231 options, 52 studies), had a median of 48 kg CO₂e/m². Using one value per study changed medians by less than 4 kg CO₂e/m² in the four largest categories and up to 72 kg CO₂e/m² in smaller ones. Therefore, small-category results are descriptive only. Comparison was limited because one-quarter of options were not reported per facade area or the reporting included stages beyond A1–A3. In addition, for 48% of options containing bio-based materials, studies did not state whether stored carbon was credited. In studies that reported both versions, the credit alone turned 15 of 26 results negative. Values should therefore be compared only within a category and at equivalent performance, boundary, and biogenic carbon treatment. To address these gaps, this review proposes a minimum reporting checklist for consistent and comparable facade LCA results.

Artificial intelligence (AI) and machine learning (ML) are promoted as tools for improving the environmental sustainability of livestock production, but it remains unknown whether the models reported in this literature are used to support real-world decisions. We systematically reviewed studies that applied AI or ML to environmental sustainability outcomes in livestock production. Four databases were searched in January 2026, and 165 of 1994 screened records were included. The protocol was registered, but the main outcome was defined after data extraction, so the analysis is exploratory. A study counted as delivering decision support only if it named a specific decision and used its model to compare options or recommend an action. The authors of 96 studies (58.2%) stated that their work could support decisions, but only 26 studies (15.8%) met our definition. This gap appeared in 10 of 11 application areas. The gap held under both a broader and a stricter definition, and when a second coder repeated the coding. Only 30 studies (18.2%) tested their model outside the setting where it was developed, and 17 (10.3%) included an optimization step, which turns a prediction into a choice. Linking ML with mechanistic farm models is a promising approach that still needs testing. We also describe the main components a practical decision tool would need.