{"version":"4.0","dims":[{"key":"problem","name":"问题与缺口","max":12,"hint":"要解决的真问题，以及为什么现在还没解决","items":[{"id":"P1","name":"问题有量化后果","w":3,"level":"minor","check":"rule","owner":"ai","criteria":"所研究的疾病或科学问题给出了可感知的规模或后果（发病率、患者数、生存率、经济损失等），不是笼统说「危害严重」。","fix":"把「严重危害健康」换成具体数字，如「我国每年新发 X 万例，5 年生存率仅 Y%」。"},{"id":"P2","name":"现状陈述有据","w":3,"level":"major","check":"rule","owner":"ai","criteria":"研究现状至少 3 条，且每条能追溯到文献（引用编号或年份），不是无出处的断言。","fix":"为每条现状补一个引用编号，或删除无法出处的断言。"},{"id":"P3","name":"缺口具象到机制/人群/环节","w":4,"level":"major","check":"llm","owner":"ai","criteria":"指出的缺口是具体的：缺哪个机制的证据、缺哪类人群的数据、缺哪个环节的验证。出现「研究较少」「尚不清楚」「有待深入」这类空话且未给出更具体的指向，即判不通过。","fix":"把「相关研究较少」改成「XX 通路在 YY 亚型中的作用尚无直接证据，现有结论均来自体外细胞系」。"},{"id":"P4","name":"缺口与假说因果衔接","w":2,"level":"major","check":"llm","owner":"ai","criteria":"从缺口推到本研究要验证的假说，是一条因果链（因为缺 A 的证据，所以要测 A），不是「顺便相关」。","fix":"补一句过渡：因为缺 X，所以本研究直接测 X，若 X 成立则 Y 应随之改变。"}]},{"key":"hypothesis","name":"假说与可证伪","max":10,"hint":"能被一次实验推翻的主张","items":[{"id":"H1","name":"有单一核心假说","w":2,"level":"blocker","check":"rule","owner":"ai","criteria":"存在一句话可陈述的核心假说。没有核心假说，整份方案就只是在做实验，不是在回答一个问题。","fix":"写一句「在<人群/模型>中，<干预/因素>通过<机制>导致<可观测结果>」。"},{"id":"H2","name":"假说三要素齐全","w":2,"level":"major","check":"llm","owner":"ai","criteria":"假说包含：谁/什么条件、发生什么变化、通过什么机制。缺任一要素都会让实验设计失去落点。","fix":"按「条件 + 变化 + 机制」补全假说。"},{"id":"H3","name":"假说可证伪","w":3,"level":"blocker","check":"llm","owner":"user","criteria":"给出了能被推翻的具体观测结果：如果测到 A 而非 B，则假说不成立。只写「验证假说成立」而不写「什么情况下不成立」，判不通过。","fix":"写清反向判据：「若干预组 X 指标未高于对照组 20%，则核心假说不成立」。"},{"id":"H4","name":"子假说 2-4 条","w":1.5,"level":"minor","check":"rule","owner":"ai","criteria":"子假说 2 至 4 条，过多说明没有收敛，过少说明没有分解。","fix":"把子假说收敛到 3 条，每条对应一个可独立验证的环节。"},{"id":"H5","name":"边界条件明确","w":1.5,"level":"minor","check":"llm","owner":"ai","criteria":"说明了假说的适用范围或不适用情形（人群、分期、模型、剂量区间），避免被质疑「过度外推」。","fix":"补一句适用边界，如「本研究结论限于 III-IV 期腺癌，不适用于鳞癌」。"}]},{"key":"chain","name":"举证链闭合","max":18,"hint":"每一环主张由哪个实验、哪个读出、什么判据支撑","items":[{"id":"C1","name":"存在显式举证链","w":4,"level":"blocker","check":"llm","owner":"ai","criteria":"方案显式给出了「主张 → 实验 → 读出 → 判据」的链条，而不是只列研究内容。评审的核心判断就是这条链是否闭合；链条缺失时无法评估方案是否可行。","fix":"按「环 1：主张 X；实验 Y；读出 Z；判据 ≥/≤ N」把链条写出来。"},{"id":"C2","name":"每环有量化判据","w":3,"level":"major","check":"rule","owner":"ai","criteria":"链条至少 3 环，且每一环的成功判据都是数字（比例、倍数、显著性、阈值），不是「明显升高」这类形容词。","fix":"为每一环补一个数字判据。"},{"id":"C3","name":"环与研究内容一一对应","w":3,"level":"major","check":"llm","owner":"ai","criteria":"每一环都能落到某项研究内容上，每项研究内容也都在链上有位置。存在孤儿内容（做了但不进链）或孤儿环节（在链上但没人做）即判不通过。","fix":"删除不进链的内容，或把它补进链的对应环节。"},{"id":"C4","name":"关键环有正交验证","w":3,"level":"major","check":"llm","owner":"ai","criteria":"最关键的 1-2 环不止一种验证手段（如表达层面 + 功能层面、体外 + 体内），避免「用同一个实验自证自己」。","fix":"为关键环补一个不同层面的验证手段。"},{"id":"C5","name":"最弱环有兜底","w":3,"level":"major","check":"llm","owner":"ai","criteria":"识别出了链上最可能断的一环，并给出了该环失败时的替代路径或降级方案。","fix":"点明最弱环，并写出失败时的 Plan B。"},{"id":"C6","name":"链终点回答假说","w":2,"level":"major","check":"llm","owner":"ai","criteria":"链条的最后一步直接回答核心假说，而不是跑到一个相邻但不同的问题上。","fix":"把最后一环的判据改成直接对应假说的表述。"}]},{"key":"method","name":"设计与方法严谨","max":16,"hint":"样本量、对照、终点、统计、偏倚控制","items":[{"id":"M1","name":"样本量有估算依据","w":3,"level":"blocker","check":"rule","owner":"user","criteria":"涉及人群或动物的研究给出了样本量及其依据（效应量、把握度、α、或引用同类研究）。没有依据的样本量是评审最常抓的硬伤，且**必须由研究者按真实参数确定**，AI 不能替他编。","fix":"按「α=0.05，把握度 0.8，预期效应量 d=0.5，双侧检验，每组需 n=64 例」写清估算。"},{"id":"M2","name":"对照与分组明确","w":2,"level":"major","check":"llm","owner":"ai","criteria":"有明确的对照组（阴性/阳性/空白）与分组逻辑，且分组能支持链上的比较。","fix":"补写分组：对照组用什么、每组多少例、如何分配。"},{"id":"M3","name":"主要终点单一且预设","w":2,"level":"major","check":"llm","owner":"ai","criteria":"预设了唯一的主要终点，其余为次要终点。列一堆平行指标会被质疑「总有一个能显著」。","fix":"指定一个主要终点，其余降为次要或探索性终点。"},{"id":"M4","name":"统计方法具体","w":2,"level":"major","check":"rule","owner":"ai","criteria":"写明了具体的统计方法（检验名称）与多重比较校正方式，而不是「用 SPSS 分析」。","fix":"写明：两组比较用 Mann-Whitney U，多组用 Kruskal-Wallis，多重比较用 Benjamini-Hochberg 校正。"},{"id":"M5","name":"偏倚控制","w":2,"level":"major","check":"llm","owner":"ai","criteria":"有针对性的偏倚控制措施（随机化、盲法、批次平衡、中心效应、批次效应校正），且与本研究的实际风险对应。","fix":"补写与本设计匹配的偏倚控制，如「按批次分层随机，检测顺序随机化」。"},{"id":"M6","name":"重复次数明确","w":2,"level":"major","check":"rule","owner":"ai","criteria":"写明了技术重复与生物学重复的次数（如 n=3 独立实验），不是「重复实验验证」。","fix":"补写：每组 3 次独立生物学重复，每次 3 个技术复孔。"},{"id":"M7","name":"伦理与数据合规","w":2,"level":"major","check":"rule","owner":"user","criteria":"涉及人体样本或动物实验的，写明了伦理审批、知情同意或动物伦理编号。缺此项形式审查即不通过。","fix":"补写伦理审批号或说明「已获 XX 医院伦理委员会批准（批号 XXX）」。"},{"id":"M8","name":"关键方法有可行性证据","w":1,"level":"minor","check":"llm","owner":"user","criteria":"关键实验方法有本单位已成功开展的证据（做过、有数据、有设备），而不是「拟采用」。","fix":"补一句「该方法本课题组已完成 X 例样本检测，成功率 Y%」。"}]},{"key":"feasibility","name":"可行性与前期基础","max":14,"hint":"前期数据、样本、平台、分工","items":[{"id":"F1","name":"有前期数据","w":5,"level":"blocker","check":"rule","owner":"user","criteria":"给出了支撑核心假说的前期数据（自己的预实验、已获得的样本检测结果），并带具体结果。**这是 AI 唯一无法代劳的一项**：编造前期数据是学术不端，只能由研究者提供真实结果。","fix":"填入真实的预实验数据；没有就写「待补：已完成 X 例预实验，预计 Y 月出结果」。"},{"id":"F2","name":"样本来源与数量落实","w":3,"level":"major","check":"rule","owner":"user","criteria":"写明了样本从哪里来、已有多少、还需收集多少、预计收集周期。","fix":"补写：本院 2024-2026 年存档蜡块约 200 例，年新增约 60 例，可满足入组需求。"},{"id":"F3","name":"平台与资质","w":2,"level":"major","check":"rule","owner":"user","criteria":"关键平台、设备或资质写到了可核验的程度（设备型号、资质名称、检测能力），不是「依托单位条件良好」。","fix":"补写具体平台：如「本院病理科配备 Illumina NextSeq 550，年检测能力 500 例」。"},{"id":"F4","name":"分工到人","w":2,"level":"major","check":"llm","owner":"ai","criteria":"任务分工落到具体角色（谁负责样本收集、谁负责数据分析），而不是「团队实力雄厚、结构合理」。","fix":"改写为：样本收集由 XX 科 A 负责，测序由 B 负责，统计由 C 负责。"},{"id":"F5","name":"进度与工作量匹配","w":2,"level":"major","check":"llm","owner":"ai","criteria":"进度安排与研究内容的工作量、周期一致，关键节点有可交付物，不是把三年平均切成三段。","fix":"按真实工作量重排节点，并给每个节点一个可交付物。"}]},{"key":"budget","name":"资源与预算匹配","max":10,"hint":"钱花在哪、依据是什么、是否超上限","items":[{"id":"B1","name":"分项合计等于总额","w":2,"level":"major","check":"rule","owner":"ai","criteria":"预算分项金额之和与总额一致（容差 0.5%）。算术错误是最容易被评审一眼看出的问题。","fix":"由系统自动校正；如为人工撰写请重新核算。"},{"id":"B2","name":"每项有测算依据","w":3,"level":"major","check":"rule","owner":"ai","criteria":"每个预算科目都有「数量 × 单价」或「按比例计提」的依据，而不是只写金额。","fix":"为每个科目补测算依据，如「200 例 × 1500 元/例 = 30 万」。"},{"id":"B3","name":"总额不超资助上限","w":2,"level":"major","check":"rule","owner":"ai","criteria":"申请总额不超过所申报项目的资助强度上限（超上限会被核减或不予受理）。","fix":"压缩到上限内，或改申报匹配的项目类型。"},{"id":"B4","name":"预算结构与设计匹配","w":2,"level":"major","check":"llm","owner":"ai","criteria":"钱主要花在研究设计的关键环节上（如大队列研究应有足够的检测与随访费用），没有与研究内容无关的大额科目。","fix":"调整科目结构，使大额支出对应链上的关键环节。"},{"id":"B5","name":"预算制式正确","w":1,"level":"minor","check":"rule","owner":"ai","criteria":"包干制项目不编明细科目（只写经费使用计划）；非包干制项目必须有明细科目。","fix":"按申报项目的预算制式调整。"}]},{"key":"defense","name":"风险与抗质疑","max":12,"hint":"最尖锐的质疑与最坏情况的处置","items":[{"id":"R1","name":"风险 ≥3 条且切题","w":2,"level":"major","check":"rule","owner":"ai","criteria":"风险分析至少 3 条，且针对本方案特有的技术/样本/时间风险，不是「实验可能失败」这类通用话。","fix":"把通用风险替换为本方案特有的风险，如「蜡块保存超 5 年导致 DNA 降解，影响 WGS 建库成功率」。"},{"id":"R2","name":"风险配最小验证","w":3,"level":"major","check":"rule","owner":"ai","criteria":"每条主要风险都配了一个最小成本的验证或规避动作（预实验、替代样本、降级方案）。","fix":"为每条风险写一个可在 1 个月内完成的最小验证。"},{"id":"R3","name":"预演评审质疑 ≥2 条","w":2,"level":"major","check":"rule","owner":"ai","criteria":"主动列出了评审最可能提出的 2 条以上尖锐质疑，并给出回应。","fix":"站在评审角度写 2-3 条最狠的质疑，并逐条回应。"},{"id":"R4","name":"有量化止损条件","w":3,"level":"major","check":"rule","owner":"ai","criteria":"给出了明确的止损条件（什么情况下停止投入），且可判定。没有止损的方案会被认为不诚实。","fix":"补写：若第 12 个月主要终点组间差异 <10% 且无趋势，则终止后续队列扩增。"},{"id":"R5","name":"负结果也有价值","w":2,"level":"minor","check":"llm","owner":"ai","criteria":"说明了假设不成立时的价值（阴性结果可发表、可否定某类干预、可修正指南）。","fix":"补一句阴性结果的出口，如「若为阴性，可否定 XX 作为该亚型标志物的价值，避免后续无效投入」。"}]},{"key":"output","name":"产出与价值","max":8,"hint":"交得出什么、谁会用","items":[{"id":"O1","name":"产出物具体可核","w":3,"level":"major","check":"rule","owner":"ai","criteria":"产出物具体到类型与数量（论文几篇、专利几项、指南/共识、产品或检测 panel），不是「发表高水平论文」。","fix":"写具体：2 篇 SCI（1 篇 IF>8）、1 项发明专利、1 个可临床转化的检测组合。"},{"id":"O2","name":"产出与研究内容对应","w":3,"level":"major","check":"llm","owner":"ai","criteria":"每项产出都能对应到某项研究内容的完成，不是凭空许诺。","fix":"把产出逐条挂到研究内容上。"},{"id":"O3","name":"有学术之外的价值","w":2,"level":"minor","check":"llm","owner":"ai","criteria":"说明了超出论文的价值：临床决策、公共卫生、产业转化、政策建议。","fix":"补一句可落地的应用场景，如「可直接用于本院肺结节良恶性鉴别，减少 X% 的穿刺」。"}]}],"total":100,"rules":"① 45 项判据逐条打分，每项 0（不满足）/ 0.5（部分）/ 1（满足）；② 21 项由程序判定，24 项由模型判定但必须摘录原文，摘录对不上的一律降级；③ 任一项阻断判据不满足，总分封顶 59，不直接归零；④ 判据标明「谁能补」——AI 能补的可以自动优化，标为「你」的只能由你提供真实材料。","disclaimer":"本评分衡量「按评审标准还有多少可被指出的短板」，不等于中标概率，也不构成可行性判断。分数高只说明能挑出来的毛病少；真实评审还看创新性的相对水平、学科竞争格局、评审人偏好与当年资助盘子。","evidence":"题库式判据参考国家自然科学基金、省级与市级课题的通行评审要点整理，非任何机构的官方评分表。"}