直接询问 AI 对论文的看法，得到的反馈往往过于宽松或缺少判断依据。
PaperJury 会先找出 reviewer 可能提出的问题，再判断这些问题是否成立；可安全修改的问题生成最小补丁，缺少实验或证据的问题交回作者，不成立的意见直接驳回。Ask an AI if your paper is any good and you get a polite yes. PaperJury argues the other side: N reviewers read the whole paper, disputes go to an independent vote and a three-way verdict (fix it / needs you / no fix), and only edits you sign off on land, then it actually compiles.
plugin 市场安装已上线（v1.0.0），下面这件事会排进下一轮。Plugin-marketplace install shipped in v1.0.0; this one moves into the next pass.
这三种模式对应不同的使用意图：局部修改使用 direct-edit，投稿前自查使用 review；需要无人值守的多轮运行时，配置 auto 策略并使用 /goal 驱动跨轮继续。Three entries, one engine. review and auto run the same courtroom engine; auto just removes the human second instance and swaps each human gate for a pre-authorized policy + return queue.
三种模式对应三种意图：系统会根据你的描述在 direct-edit 与 review 之间选择；auto 不会自动启用。无人值守的多轮运行需要配置 auto 策略，并使用 /goal 驱动跨轮继续。The three modes are not three commands but three intents: direct-edit and review are auto-routed by what is said; auto is the only explicit switch (/goal) and is never self-detected.
以下展示一条问题从提出到应用修改或交回作者的完整路径。机械性或轻微问题不会进入庭审，而是进入单独的快速 polish 流程；只有重大且有争议的问题才会进入裁定。Below is the order the engine processes one issue, step by step; mechanical and minor issues skip the trial and take a separate quick polish path.
将内容分解为 claim 单元，并保留稳定的 passage-id。Split into claim units + stable passage-id
decompose.js为论文确定 N 个子领域，并由 N 位领域评审各自通读全文一遍。assign N subfields; N domain reviewers each read the whole paper once
assign-reviewers · reading-check执行防略读审计和跨评审去重，然后按争议程度进行分流。anti-skim audit + cross-reviewer dedup → route by contestability
coverage-auditor · merge5 人陪审团基于局部上下文投票；无明显多数时通常增加到 12 人，全员都因上下文不足而无法判断时直接交回作者。代码判定指控是否成立，judge agent 再路由已成立的指控。whole-paper defense / 5-juror local-context tier / escalate to 12 → judge routes three ways
trialMode A 找回被误判为不成立的问题；Mode B 在应用修改前抽检强共识重大问题。Mode A revives wrong drops + Mode B spot-checks consensus majors before the edit
recall-audit对于可安全修复的问题（valid-fixable），起草最小补丁。valid-fixable → a minimal patch
drafteranchor-diff 和 cross-ref 先预筛风险；RISKY 修改还要经过语义审计。随后由 apply-patch 原子应用，再运行 compile-guard，失败就回滚。anchor-diff + cross-ref → meaning / edit-audit → apply
anchor-diff · cross-ref · apply-patch在每轮结束时核对残留问题并判断是否收敛；review 模式由作者把关，auto 模式由 clerk 收尾。round-boundary reconcile + converge; review = the author's gate, auto = clerk converges
clerk / queue每条问题单独处理：提出问题的评审只负责找出问题，不参与裁定；独立评审组查看双方证据后投票，代码按照法定人数和多数规则判定指控成立或不成立。对于已成立的指控，judge agent 再决定是安全修复还是交回作者。下图沿用法庭隐喻，将检方、辩方、陪审团、法官和作者置于各自的位置。Each issue = one case; the defendant = the passage under attack, not the issue itself. Position is stance: prosecution left, defense right, jury center, judge above, appeal outside.
| 法庭角色Courtroom role | 引擎实体Engine entity | 立场 / 职责Stance / duty | ⚠ 别混淆⚠ Do not confuse |
|---|---|---|---|
| 指控（charge）charge | 一条问题one issue | 受审的「案由」the matter on trial | 是「指控」，不是被告，也不是 AI 角色it is the charge, not the defendant, not an agent |
| 被告defendant | 受攻击的 passage / claimthe passage / claim under attack | 被动，不发言passive, does not speak | 不是 AI 角色，也不是问题本身not an agent, not an issue |
| 检方prosecution | N 位领域评审（每位通读全文）N domain reviewers (each reads the whole paper) | 提出指控并附引文files charge + quote | 提出后退场，不裁定，也不为自己提出的问题辩护steps away after filing, does not judge / does not defend its own issue |
| 辩方defense | author agent（作者侧代理）author agent | 带证据为其辩护：已处理 / 越界 / 修改会导致漂移argues its case with evidence: already addressed / out of scope / fixing it would drift | 不是人类作者；之后转为起草器（drafter）= author AGENT, not the human; later becomes the drafter |
| 覆盖 + 合并coverage + merge | coverage-auditor + merge（workflow）coverage-auditor + merge (workflows) | 防止略读的审计 + 跨评审去重 + 按争议程度分流anti-skim audit + cross-reviewer dedup + route by contestability | 机械 / 轻微问题走 polish 流程，不进庭审mechanical / minor items take the polish track, not the trial |
| 庭审陪审团trial jury | 先 5 位、必要时升到 12 位、视角彼此独立的全新陪审员5, escalating to 12, fresh jurors with independent perspectives | 中立，带着局部上下文、依据双方证据投票neutral, votes on both sides' evidence with local context | 既非领域评审也非 author agent；无明显多数时通常升到 12 人，全员 context-limited 时不升级fresh, neither reviewer nor author; escalates to 12 only on no clear majority |
| 法官judge | 1 个 judge agent1 presiding agent | 在指控成立后路由为 valid-fixable 或 author-required，并设定 criterionpresides, tallies, rules, sets criterion | 不参与投票，也不改变多数票结果rules only, does not vote |
| 起草器（drafter）drafter | author agent（转换角色）author agent (different stage) | 对可安全修复的问题（valid-fixable）起草最小改动；需要新数据就如实软化或交回作者minimal-edit for valid-fixable; for needs-data, honest softening or escalate | 同 persona、不同阶段same persona, different stage |
| 召回审计recall audit | 新的怀疑者（fresh skeptic）fresh skeptic | Mode A 复查每条驳回件；Mode B 在应用修改前抽检强共识重大问题Mode A re-checks every drop; Mode B spot-checks consensus majors before the edit | 不是原评审（避免固守先前判断）not the original reviewer (avoids entrenched positions) |
| 二审 / 终审second / final | 人类作者the human author | 只看升级件，终裁记录在案sees only escalations, final ruling logged | = 人，不是 author agent= the human, not the author agent |
| 执行官 / orchestratorbailiff / orchestrator | orchestrator + 确定性脚本orchestrator + deterministic scripts | 运行护栏 / 维护 ledger / 写 journalruns guards / keeps ledger / writes journal | 庭外，不裁定，只运行确定性检查和收敛检查off-court, does not judge, only runs the deterministic guards + gate |
| 书记官 clerkclerk | clerk(workflow)clerk (workflow) | 轮边界：核对遗留问题 vs 本轮改动、去重、判定是否收敛round boundary: reconcile carried open-questions vs this round's edits, dedup, decide convergence | 语义 workflow，不是上面的法警（确定性脚本）a semantic workflow, not the bailiff (deterministic scripts) above |
分工原则是：召回靠阅读，精度靠裁定。阅读阶段只负责找全问题并提供引文，不判断问题是否成立；每条观察都必须引用原文，无法提供引文说明评审没有充分阅读，也可能是在编造。N 位领域评审各自通读全文一遍，再通过三层机制减少略读。Organizing principle: recall belongs to reading, precision to the courtroom. Readers only find + cite, they do not judge validity; they must quote the source (cannot quote = did not read, anti-skim and anti-hallucination). N domain reviewers each read the whole paper once, backstopped by a three-layer anti-skim.
PaperJury 不默认相信评审意见。每条问题先核验，再按性质分流：能安全修改的写补丁；缺少实验、证据或需要研究判断的交给作者；不成立的直接驳回。Routes in two stages: first a deterministic contestability router sends mechanical + minor issues to a separate polish track (no jury); only substantive-major issues reach the trial. The three routes below are the judge's verdict inside the trial, downstream of the contestability router.
误读、幻觉、重复、越界或严重度虚高：驳回，并保留证据。accurately judged invalid (hallucination / misread / duplicate / out of scope / inflated severity) → dropped, with evidence.
→ 仍会交给召回审计抽查→ to recall audit for re-check这类问题包括文本表达、claim 过强或结构不清，不需要新增实验或数据，也不会改变原意。系统会生成最小补丁，通过护栏检查后再应用。valid + text-fixable (no new experiments/data) + no drift → draft a minimal edit, pass the guards, land it.
交给起草器（drafter）处理；如果需要新数据，则改判为作者处理（author-required），或在不夸大结论的前提下软化表述。→ drafter · if it needs new data, reroute to author-required or soften honestly缺少实验、数据，或需要研究判断：交回作者决定。这是正确分流，不算漏判。needs author-private info / new experiments / a judgment call → routed to the author (second instance). Correct handling, not a recall loss.
→ review 交作者 · auto 放入待办→ review hands it to the author · auto queues it