Silicon increases wheat grain yield under field stress: reproducibility package (data, code, prompts, audit ledgers)

Reproducibility package supporting the manuscript "Silicon increases wheat grain yield under field stress: a human-reviewed, LLM-assisted meta-analysis" (Halpern, 2026; submitted to Agronomy for Sustainable Development). Contents: final extraction datasets, analysis inputs, dual-pipeline LLM extraction outputs, contrast-rescue ledger, audit ledgers and citation audit, prompt inventory and run metadata, analysis scripts, and the rebuilt manuscript review draft. Source PDFs of the included primary studies are excluded from this public package for copyright reasons; a separate internal audit package retaining the PDFs is held privately for traceability and will not be redistributed. Primary result: silicon increased wheat grain yield by +11.35% (95% CI +7.49% to +15.34%; 257 comparisons from 30 studies) under field conditions, with the largest effects under salinity (+36.0%, k=18, 4 studies) and drought (+13.4%, k=58, 9 studies). The estimator is a three-level method-of-moments random-effects model with CR2 cluster-robust variance and Tipton-HTZ degrees of freedom; details and reproduction scripts are inside the package. Companion preprint: https://doi.org/10.64898/2026.02.17.706322.本可复现性包配套支持论文《硅在田间胁迫下提升小麦籽粒产量：一项人工审阅、大语言模型（LLM）辅助的元分析》（哈尔彭，2026年；已提交至《Agronomy for Sustainable Development》（可持续发展农学））。 本包包含以下内容：最终提取数据集、分析输入文件、双流水线大语言模型提取结果、对照校正台账、审核台账与引用审核文件、提示词清单与运行元数据、分析脚本以及修订后的稿件评审草案。因版权限制，本公开包未纳入所收录原始研究的源PDF文件；另有一份留存了PDF文件的内部审核专用包用于可追溯性目的，将不进行再分发。 核心研究结果表明，在田间胁迫条件下，硅可使小麦籽粒产量提升11.35%（95%置信区间：7.49%~15.34%；共包含30项研究的257组对比数据）；其中盐胁迫下增效效果最为显著（提升36.0%，k=18，涉及4项研究），干旱胁迫下增效为13.4%（k=58，涉及9项研究）。本研究采用的估计方法为三级矩法随机效应模型，搭配CR2聚类稳健方差与Tipton-HTZ自由度校正；详细方法与复现脚本均包含在本包内。 配套预印本：https://doi.org/10.64898/2026.02.17.706322