project_manifest.json |
项目名称、PRISMA 版本、AI 模式、应用版本等基本信息Project name, PRISMA version, AI mode, app version and other metadata |
events.jsonl |
按时间顺序记录导入、去重、筛选、复核、质量评价和导出的所有事件Chronological log of all import, dedup, screening, review, quality, and export events |
screening_decisions.csv |
每条文献的筛选阶段、纳入/排除决定、排除理由、审稿人、冲突状态和 AI 使用情况Per-record screening stage, include/exclude decision, exclusion reason, reviewer, conflict status, and AI usage |
exclusion_reasons.csv |
标准排除理由分类和每类计数，适合放在 PRISMA 图下方Standardized exclusion reason categories and counts, suitable for PRISMA diagram annotation |
prisma_counts.json |
从审计数据重新计算的 PRISMA 流程图数字，可用于准备 PRISMA 2020 checklist，仍需研究者核对PRISMA flow-diagram counts recalculated from audit data to support PRISMA 2020 checklist preparation; researcher verification remains required |
quality_appraisal.csv |
质量评价导出：每条纳入研究的模板、工具族、领域判断、支持引文、审稿备注和状态Quality appraisal export: template, tool family, domain judgement, supporting quote, reviewer note, and status for included studies |
evidence_table.csv |
证据表导出：每条纳入研究的 PICOS、效应量、质量判断、证据确定性和备注；GRADE 仍需人工确认Evidence table export: PICOS, effect measure, quality judgement, certainty of evidence, and notes for each included study; GRADE remains human-confirmed |
grade_summary.csv |
GRADE 摘要骨架：按 outcome / PICOS 聚合研究、效应摘要、质量判断摘要和基线确定性；最终 GRADE 和降级理由保留给人工确认GRADE summary scaffold: groups studies by outcome / PICOS with effect summary, quality judgement summary, and baseline certainty; final GRADE and downgrade reasons stay human-confirmed |
audit_summary.md |
可读的审计摘要：事件数量、决策数量、PRISMA 计数、AI 模式、数据边界和风险提示Human-readable audit summary: event counts, decision counts, PRISMA counts, AI mode, data boundary, and risk notes |
DEFENSE_AUDIT_PACK.md |
方法附录 / 复核证据包：整合 PRISMA 计数、双审冲突、一致性指标、质量评价、中文源可靠性提醒和 AI 边界说明；使用前需由研究者核对Methods appendix / review evidence package: combines PRISMA counts, dual-review conflict status, agreement metrics, quality appraisal, Chinese-source reliability warnings, and AI boundary notes; researcher verification is required before use. |
export_snapshot_manifest.json |
M5 导出快照清单：列出证据包文件、媒体类型、字节数、SHA-256 hash、producer 和 manifest hash，用于复核本地导出集合是否一致。M5 export snapshot manifest: lists evidence-package files, media types, byte counts, SHA-256 hashes, producer, and manifest hash so a local export set can be checked for consistency. |
ai_usage_registry.json |
AI 模式、服务边界、允许阶段、数据边界和用户确认记录；这是配置证据，不是最终筛选决定AI mode, provider, allowed stages, data boundary, and user acknowledgement log; configuration evidence, not a final decision ledger |
ai_suggestions.jsonl |
AI 建议、理由、置信度、输入指纹、人工处理结果、复核时间、人工改写字段、关联的人类筛选决定和计数边界；被拒绝的建议不会进入 PRISMA 计数AI suggestions, rationale, confidence, input hashes, human handling outcomes, reviewed_at, human edit fields, linked human decisions, and prisma_count_boundary; rejected suggestions do not enter PRISMA counts |
PRISMA_TRAICE_REPORT.md |
PRISMA-trAIce 透明报告：说明是否使用 AI、如何使用、No-AI 状态，以及“最终计数来自人工 ScreeningDecision”的边界PRISMA-trAIce transparency report covering AI usage, No-AI state, and the boundary that final counts come from human ScreeningDecision records |
dual_review_conflicts.csv |
V2.5 dual-review evidence: screening and quality conflicts, reviewer A/B decisions, resolver final decisions, and statusV2.5 dual-review evidence: screening and quality conflicts, reviewer A/B decisions, resolver final decisions, and status |
dual_review_agreement.json |
V2.5 agreement metrics: percent agreement, Cohen's kappa, paired decisions, and unresolved-conflict export gate statusV2.5 agreement metrics: percent agreement, Cohen's kappa, paired decisions, and unresolved-conflict export gate status |