这是一张我们对家长、复核人和工程师都用同一份口径的表。原则：规则与外部数据能坐实的才叫"确认"或"判断"；其余一律以"招生官会怎么问"的形式提示，不下结论。The same table we give parents, reviewers and engineers. The principle: only what rules or external data can establish counts as confirmed or determined. Everything else is raised as a question a reviewer would ask — never as a conclusion.
能正面确认Can be positively confirmed
核实成立、可写进报告的结论——不只是找问题Established facts we can state in the report — not only problems
文件自签发起未被改动Document unaltered since it was issued数字签名校验通过且覆盖全文。官方电子成绩单与分数报告普遍带此类签名。注意语义：能证明的是"未被改动"，不是"内容为真"Digital signature valid and covering the whole file — standard on official electronic transcripts and score reports. Note the limit: this proves the file was not altered, not that its contents are true
荣誉已在主办方公开名单核实Honor confirmed against the organizer’s public list申请人姓名命中名单且申报奖级与名单一致；同名不可消歧时不确认The applicant’s name appears with a matching award tier; never confirmed when namesakes cannot be told apart
声称的论文或代码已按标识符核实A claimed paper or repository verified by its identifierDOI／arXiv／GitHub／ORCID 直连公开接口取回记录，在本地比对作者列表——申请人姓名从不外发，第三方只拿到一个标识符The record is fetched straight from the public DOI, arXiv, GitHub or ORCID interface and the author list is matched locally — the applicant’s name is never sent out; the third party sees only an identifier
竞赛与夏校的官方通道现场查得Found through the organizer’s official channel只认主办方官方域名的页面作为证据，并在本地复核姓名与年份Only pages on the organizer’s own domain count as evidence, re-checked locally for name and year
能判断Can be established
算术、公开规则或外部名单可直接坐实的矛盾Contradictions that arithmetic, published rules, or an external list settle directly
标化成绩取值越界Test score outside its valid rangeSAT 400–1600 且为 10 的倍数、ACT 1–36、AP 1–5、IB 单科 1–7／总分 ≤45、TOEFL 0–120、IELTS 0–9 且为 0.5 的倍数、Duolingo 10–160 且为 5 的倍数——出题方公开规则SAT 400–1600 in steps of 10, ACT 1–36, AP 1–5, IB 1–7 per subject / 45 total, TOEFL 0–120, IELTS 0–9 in steps of 0.5, Duolingo 10–160 in steps of 5 — all published by the test makers
分项之和与总分不符Section scores do not sum to the total算术Arithmetic
课程体系学制矛盾Course level impossible for that gradeIB Diploma 为两年制，只在 11–12 年级；A-Level 对应 Y12–13；IGCSE 对应 Y10–11The IB Diploma runs two years (grades 11–12); A-Level maps to Y12–13; IGCSE to Y10–11
GPA 与排名的算术矛盾GPA or class rank does not add up未加权 GPA 不会超过满分；名次不会大于年级总人数An unweighted GPA cannot exceed the scale; a rank cannot exceed the class size
日期倒置与同期时长超载Reversed dates and impossible workloads结束早于开始；按月聚合 >60h/周 提示、>80h/周 强信号；学期中 ≥30h/周 与全日制在校身份冲突End before start; aggregated monthly at >60h/wk flagged and >80h/wk strong; ≥30h/wk during term conflicts with full-time enrollment
荣誉年份与年级、毕业年份矛盾Honor year contradicts grade level or graduation year以毕业年份为锚换算年级对应学年；获奖年份晚于毕业年份即矛盾Grade levels are anchored to the graduation year; an award dated after graduation is a contradiction
奖项名称与填报级别矛盾Award name contradicts the level claimed名称字面级别 ≠ 填报级别；名称推断不出级别则不出结论The literal level in the name vs. the level claimed; silent when the name implies none
Common App 硬约束超限Common App hard limits exceeded活动 >10、荣誉 >5、字段与文书字数超限——说明手上不是最终填报版>10 activities, >5 honors, over-length fields or essays — meaning this is not the final submitted version
签名无效或签署后被追加修订Signature invalid, or content added after signing签名校验未通过，或有效但未覆盖全文——均为文件结构事实Signature fails validation, or is valid but does not cover the whole file — both are structural facts
给出的论文标识符无法解析A cited paper identifier does not resolveDOI 由注册机构发放、是唯一键——doi.org 与 Crossref 均查无此号即为硬事实；arXiv 编号同理A DOI is a unique key issued by a registration agency: if neither doi.org nor Crossref knows it, that is a hard fact. The same holds for arXiv IDs
推荐信的作者元数据为申请人本人Recommendation letter authored by the applicantPDF 的 Author 字段与申请人姓名一致（含拼音倒序）。核查方最常用的一招The PDF’s Author field matches the applicant’s name, including reversed pinyin — one of the first things a reviewer checks
只能提示风险Raised as a question only
以追问形式呈现，需人工复核，绝不下结论Presented as a question, human-reviewed, never a conclusion
课程 ↔ 科研 ↔ 竞赛的逻辑匹配Coursework ↔ research ↔ competition coherence按前置／级别／深度／角色／内部矛盾五类出招生官式追问Reviewer-style questions across prerequisites, level, depth, role, and internal consistency
证据链缺失Missing evidence chain证明人／可验证产出／时间戳痕迹三类全缺时提示Flagged when a referee, a verifiable output, and a timestamped trail are all absent
文书声称在申请表里找不到对应条目An essay claim with no matching entry on the application逐条对账活动、荣誉、夏校、经历与证明材料；字面重合极低才提示Each claim is reconciled against activities, honors, programs, experiences and supporting documents; flagged only on very low overlap
声称的产出拿不出可查标识符A claimed output with no verifiable identifier声称发表论文却全包无 DOI／arXiv 编号／ISBN；声称开源却无仓库地址A publication claimed with no DOI, arXiv ID or ISBN anywhere; open-source work claimed with no repository
推荐信的人物一致性Recommender consistency亲属关系；自述任教学科在成绩单里没有对应课程；推荐人无法联系Family relationships; a subject the recommender claims to teach with no matching course on the transcript; a recommender who cannot be reached
文档来源线索Document provenance signals机构类文件出自设计软件、修改时间晚于创建、多代增量保存、经 OCR 转录——恒为提示级，正常场景也会触发Institutional files produced in design software, modified long after creation, saved across many revisions, or transcribed by OCR — always advisory; ordinary workflows trigger these too
荣誉无法核验Honor not verifiable名单未覆盖或无姓名，明确不代表不真实The list does not cover it, or no name is available — this does not imply the record is untrue
夏校选拔性口径错位Summer program described more selectively than it is付费开放项目被描述为选拔性，或免费选拔项目出现付费表述An open-enrollment paid program described as selective, or a free selective program described as paid
第三方机构痕迹Third-party agency traces材料中出现机构信息。机构协助整理完全正常，提示的是"哪些部分是你本人完成的"这条界线Agency details appear in the materials. Working with an agency is entirely normal — the question is only where your own work ends
模拟追问演练Mock interrogation drill供学生自测，不构成任何指控For the student’s own rehearsal; not an allegation of any kind
不做Not done
有些是做不到，有些是我们主动选择不做Some we cannot do; some we choose not to
AI 生成文本检测 —— 主动不做AI-writing detection — a deliberate choice not to现有方法的准确率支撑不了任何结论，而误判会直接伤害诚实的学生——那正是我们的客户。这一条不是能力不足，是我们选择不做No current method is accurate enough to support a conclusion, and a false positive lands squarely on an honest student — exactly the person we serve. This is a choice, not a limitation
文风判定"是否本人所写"Judging authorship from writing style写作风格随时间、题材、修改轮次自然漂移，把它当证据不负责任Style drifts naturally with time, topic and revision; treating it as evidence would be irresponsible
印章、手写签名的视觉鉴定Visual authentication of seals and signatures需要出具方样本库，且错判代价极高。无数字签名的文件我们不判真伪It needs a reference library from each issuer, and the cost of being wrong is severe. Without a digital signature we make no authenticity call
中国高中成绩单与校方系统核对Checking a Chinese high-school transcript against the school’s own system无公开数据源。美国院校的学分类成绩单可经 Parchment／National Student Clearinghouse 等机构通道验真No public data source exists. Credit-bearing U.S. transcripts can be verified through Parchment or the National Student Clearinghouse
打电话／发邮件联系推荐人核实Contacting recommenders to verify in person属人工服务范畴，不在自动化管线内A human service, outside the automated pipeline