PaperConan论文柯南

@user_a26c8b8f/paperconan

给论文补充材料里的源数据表格做"数值体检"的开源命令行工具 + AI 技能。它用确定性检测器扫描 .xlsx / .xls / .csv / .tsv 以及 PDF、Word 里的表格，标出值得人工复核的位置：跨列恒定倍数或偏移、跨表复制、行向复用、共享小数尾、GRIM/GRIMMER 均值一致性、末位数字分布等 30 余种统计信号，精确到文件、sheet、行列和触发规则。重要：所有结果都是"统计信号，不是结论"。工具只报告位置和模式，不判断动机，也不指控任何人；最终解释需要原始数据、图注、Methods 与作者说明。安装（需要 Python 3.10+）：pip install "paperconan[all]"。如果 pip 报 externally-managed-environment（PEP 668，常见于 Homebrew / Debian 自带的 Python），改用 uv tool install "paperconan[all]" 或 pipx install "paperconan[all]"。使用：1) paperconan 目录 —— 扫描一个源数据目录，输出 audit/scan.json 与可视化 report.html；2) paperconan fetch "DOI或标题" —— 按 DOI 查找并下载公开补充数据（0.8.4 起恢复 Nature 系补充材料抓取）；3) paperconan overview / drill / explain —— 由粗到细分层解读扫描结果；4) 配合本技能，AI 按内置规程（分层读取、误报甄别、中立措辞）给出复核建议与解读报告。源码与完整文档：https://github.com/zixixr/paperconan — English — PaperConan runs a battery of deterministic numeric detectors over a paper's supplementary source data (.xlsx / .xls / .csv / .tsv, plus tables inside PDF and Word files) and surfaces the exact locations worth a human re-check: cross-column constant ratios and offsets, cross-sheet duplication, row reuse, shared decimal tails, GRIM/GRIMMER mean consistency, last-digit distributions and more — 30+ statistical signals, pinpointed to file, sheet, rows and rule. Every hit is a statistical signal, never a verdict: the tool reports locations and patterns, not intent. Install (Python 3.10+): pip install "paperconan[all]"; on PEP 668 externally-managed systems use uv tool install "paperconan[all]" or pipx install "paperconan[all]". Usage: paperconan

scans a source-data directory into audit/scan.json + report.html; paperconan fetch "" finds and downloads open supplementary data (Nature supplement fetching restored in 0.8.4); overview / drill / explain read the results in layers; with this skill loaded, the AI follows the built-in protocol (layered reading, false-positive screening, neutral language) to produce a review-ready interpretation. Source & docs: https://github.com/zixixr/paperconan

v1.0.2