覆盖什么

在任何人从技术上评估你的工作之前发生的那些轮次。具体说就是招聘人员筛选轮、人力资源筛选轮，以及和用人经理、创始人或者 PI 的第一次谈话。边界由谁在这通电话上划定，不由他们问什么划定。用人经理在一次介绍性通话里问技术问题，仍然属于本卷；一场安排好的技术评估不属于。

创始人与 PI 那一轮单独成章，因为证据显示它运行在另一套逻辑上，可预测性远低于其他任何一轮。

不覆盖：编程题、带回家作业、系统设计、研究深挖。那些属于技术卷。

材料从哪来

大约一千五百份第一手记述，专为本卷收集：招聘人员和用人经理描述自己的做法，候选人描述自己遇到了什么，PI 和创始人描述他们怎么决定，以及在系统机制本身就是问题时采用的厂商文档。

泛泛的职业建议站点是按规则排除的，不是凭判断排除的。搜索引擎在这个题目上返回的东西几乎都是为排名而写，不是为告知而写，把它们收进来只会复制一遍本卷本来就要拿去检验的那些建议。

正文不标注来源。每一条主张在收集时都做过核验，带链接、日期和发言人身份的底层记录保存在本地。正文里出现的是强度标记，它告诉你一条主张能承受多大重量。

证据标记

● 多个独立的第一手来源一致。可以据此行动。

◆ 单一第一手来源，或者来自同一个地方的一小组来源。大概率为真，值得知道，不值得为它重排结构。

◇ 从旁证推断，没有直接佐证。是推理，并且明确标为推理。

看上去一致不等于独立一致。有一份写于 2024 年的记述，被发现在 2026 年以另一个名字逐字重发。标记反映的是独立来源的数量，不是某件事被重复了多少次。

四条赛道

最强的单项发现是：第一通电话不是一件事。存在四种不同的流程，在其中一种里有用的建议，在另一种里有害。

A. 小型研究组织。 一家 AI for science 创业公司、一家小型生物技术公司、一家非营利机构。往往完全没有非技术层；由一位科学家主持第一通电话，几分钟内就转入技术。

B. 大药企。 一位远程的、非技术的招聘专员掌管整道闸。行为胜任力打分。刚性薪档。以月为单位的时钟。

C. 研究机构、国家实验室、博后。 科学之前没有人力资源。由 PI 主持。你被期待带着一个方向到场。

D. 大型科技公司。 所有标准建议都在描述的那种结构化招聘人员筛选轮。

答案随赛道不同的地方，会标出赛道。不随赛道变的地方，不标。

本卷不会做的事

它不会靠站到更流行的一边来消解矛盾。第一手记述真正分歧的地方，两边都给出，并在其中一方站得住时给出分歧的归因。那份登记表本身是独立的一章，也是这本书里诚实的部分。

规则

每家公司投两到五个岗位，不是一个。 每家只投一个，比证据里任何一位招聘人员建议的都保守得多。红旗是不连贯，不是数量。一位招聘人员把真实门槛放在 “well over 100 roles… zero business applying for”。把这一组岗位控制在同一个职位族和同一个资历档里，并在同一个短窗口内提交。●

主动告诉第一个联系你的招聘人员，你还投了别的。 这是一句话的事，它消除了招聘人员真正点名的唯一一种失败模式：每个岗位编制各有一位招聘人员时，“one thinks the other is working with you and moves on, but then no one ends up moving forward.”。●

一个邮箱，一个电话，一个主页链接，永远不变。 申请追踪系统（ATS）不按邮箱去重。有一个系统按姓名、电话和主页链接的组合来匹配。另一个按姓名、城市和国家匹配，完全不看邮箱。所以第二个身份藏不住任何东西，还让你失去控制：档案后来会被合并，而在某一个系统里，执行合并的管理员 “won’t get to choose which one is retained”。●

要点随便改。职位名称、雇主、日期绝不能改。 旧版本会留在系统里：“our ATS does have a record of your old resumes.”。日期是雇佣核实真正会去查的字段。有一个有记录的案例：候选人如实修正了一个职位名称，旧版本还在档案里，流程立刻终止。在一个大型系统里，你最近一次申请还会自动填充之后的每一次申请，于是最新版本悄悄变成你在那里的身份。●

早投，并且不要为一家大雇主花四十五分钟做定制。 早投的优势是真的，但集中在头一两周，并且在小公司最强，那里的用人经理描述他们会锁定一份四五个人的短名单，再也不回头看。在批量评审的雇主那里，等到有人来看时，到达顺序已经看不见了。证据里的两派，主张走量的一派和主张深度定制的一派，都认为在大雇主那里逐份深度定制回报很差。●

直接开口要内推，同时继续投。 一项针对某个大雇主的研究测出的效应，是拿到面试的概率增加约七个百分点，拿到 offer 的概率增加约两个百分点。它买到的是入口，不是背书。先投递、之后再拿到内推，并不浪费。弱关系有用。点明共同的关系，把真正的请求提出来；不要以想聊聊了解情况开场。●

主动有害

• 第二个邮箱或第二个账号。会在别的字段上被识别出来，之后被合并，你会失去对哪一个版本的你能留下来的控制。●
• 在不同简历版本之间改动职位名称、雇主或日期。要点是免费的，这些不是。●
• 在一家公司的冷却期内重新申请。至少有一个主流系统提供可配置的规则，自动拒绝并把申请标记为封锁，全程没有任何人看到。真实面试流程之后按大约六个月计。没有面试的书面拒信不带这种惩罚。●
• 在同一家公司申请跨越不同资历层级或不相关职能的岗位。●
• 在担保问题上答得不准确。它是一道软过滤，不是自动拒绝：“it is not an immediate no… it does lower your chances”。它还会在 offer 阶段被核实，那是被抓到的最糟时点。●
• 给陌生人发含糊的求聊消息。证据里没有人反对的那个版本不一样：联系一位周边的人，问清楚谁负责这个流程，然后自己去找那个人。●

证据没有定论的一件事

不是真实空缺的招聘帖是否普遍。一家厂商给出接近百分之二十的数字；一家政府研究服务机构表示没有官方统计，并指出发布这些数字的厂商本身在卖招聘服务；企业内的招聘人员否认这么做。可以核实的是常青岗位编制，按厂商自己的文档，它 “no relationship to positions”，也不能直接把你招进来。投递之前没有办法把它认出来，所以实际的做法是在第一通电话里问招聘时间表。

通话之前怎么判断你在哪条赛道

读邀约，不要读职位名称。四样东西会暴露它。

谁发的，他的头衔是什么。 人才合作伙伴、招聘专员或者人力运营协调员，意味着存在一个专门的筛选层。一位具名的科学家、一位课题组长、一位研究负责人或者创始人，意味着不存在。●

时段有多长。 二十五到三十分钟加一个预约链接，是结构化筛选轮。四十五到六十分钟并且由一位科学家直接约的，是一场披着友好名字的学术对话。●

邀约有没有点明下一步。 结构化流程会预先讲清顺序。小型组织和研究组通常不讲，因为还没定下来。◆

招聘帖挂在哪里。 挂在大型招聘门户上，带岗位编制号、带内部职级、带公开的宽薪档，是一家分档组织。挂在小型申请追踪系统上，薪档很窄或者没有薪档，就不是。●

赛道 A. 小型研究组织

这里指一家 AI for science 创业公司、一家小型生物技术公司，或者一家非营利研究机构。你大部分的投递都落在这里，也是标准建议失效最严重的一条赛道。

常常根本没有非技术层。 在一家机构里，内部招聘人员拿的是生物学博士学位。在另一家，第一次谈话由招聘团队自己进行，人力资源之后才出现。在国家实验室的课题组里，用人经理和课题组长自己主持那三十分钟的筛选轮。没有热身轮。●

第一通电话考的是研究判断力，不是事务性条件。 一位候选人把某家公司的电话筛选轮描述为 “trying to assess my research ability, my critical thinking ability and my areas of expertise”，不是过一遍简历。◆

一个开放的、没有定义的问题会立刻抛过来，被测量的是速度。 有一通二十分钟的电话，结尾是一份书面作业：为某个疾病设计一次筛选。一位在小公司里主持博士候选人面试的工程师说，他宁愿招 “someone who comes up with a sub-par solution quickly and then iterates”。安静下来先想的本能在这里是错的本能。出声给一个粗糙的答案，再当着他们的面改进它。●

研究报告来得早。 在一家公司，一场二十分钟的报告就是第二轮。小型组织把研讨报告前置；大型组织留到最后。◆

薪档要么在第一通电话里当面说出来，要么根本不存在。 有一家机构的人力资源联系人在第一通电话里就报了一个区间。一家四个人的公司完全没有薪档，而股权那一半没法定价，因为 “options are worthless until something happens later”。薪酬直接跟着融资阶段走。●

这通电话是你唯一一次给这家机构能不能活下去定价的机会。 现金跑道、临床试验是不是真的注册了、创始人能不能答上一个关于科学的硬问题。在大雇主那里没有对等的信号可拿。●

担保在这里更差，不是更好。 非营利机构的 cap-exempt 身份被描述为 “almost impossible after research grant cuts”。不要以为非营利就等于对签证友好。◆

赛道 B. 大药企

薪酬是开头几分钟的硬闸，不是收尾时的谈判。 薪档按行业薪酬调查定价，并且双向执行：“most, if not all, major biotech/pharma companies have a policy to not pay outside of brackets be it above or below”。有一位候选人报了数字之后立刻被刷。保护性的那一半是真的：报一个低数字不会把你压到薪档下限以下。危险的那一半也是真的：报一个高于薪档的数字，事情就结束了。●

第一轮是行为胜任力打分，不是技术评估。 有一家大公司把面试问题公开出来，并对应到它宣称的价值观上。另一家用书面形式要求候选人按情境、任务、行动、结果的格式作答。一位候选人的总结是 “It’s every big pharma I’ve interviewed at.”。这个阶段没有编程题，也没有带回家作业。如果你准备的是技术深度，没有准备结构化的行为题答案，你准备的是另一轮。●

做筛选的人是远程的，经常在另一个国家，非技术出身，并且掌管整道闸。 一位用人经理讲得很直白：“a global remote team of HR recruiters do all the front end resume screening and phone screenings. Hiring managers may only ever see 10-15% of applicants resumes.”。除非这个人放你过去，否则你关于方法质量说的任何话都到不了科学家那里。●

时钟以月计，沉默是结构性的。 光是约上第一通电话就要三周或更久。从筛选轮到用人经理十到十四天。某家大公司一个计算生物学岗位的汇总数字约为一百二十天。一位两边都经历过的候选人说 “biotech companies tend to move faster than pharma… I onboarded at said biotech 5 weeks from the initial screening call.”。这里的沉默不带评价含义。●

担保在任何人跟你说话之前就已经定了，而且它能推翻用人经理。 “We screen for visas via HR/ recruiter prior to me even seeing the resumes.”。有一位经理干脆不再考虑任何需要担保的候选人，因为人力资源反复卡掉已经通过他自己面试的人。●

内推买到的是注意力，别的什么也没有。 “all a referral really does is ensure a human looks at your resume”。“Referral increases your chances of being interviewed, not your chance of being hired.”。有几位内部人士描述过那种为了让一个已经内定的人显得合规而存在的招聘帖，通常是一位内部候选人，或者一位正在转成正式岗的博后。●

赛道 C. 研究机构、国家实验室、博后

科学在人力资源之前，如果人力资源会出现的话。 一位在国家实验室的候选人写道 “I wasn’t first interviewed by an HR type person before being invited for an interview (I assumed this is standard practice).”。在另一处，预筛由 “host PI and core project members” 主持，人力资源只在 offer 阶段出现，来谈薪水。●

技术追问从第一分钟就开始，电话也可能就在那里结束。 一位刚做完三十场博后面试的 PI：“I started asking these questions at the start of the interview. If during this technical portion of the interview it becomes apparent a candidate has lied about the expertise I end the interview early.”。◆

做功课是一道闸，不是加分项。 多位课题组长直接丢掉泛泛的联系。一位负责跟候选人谈话的实验室成员报告说 “half the time the interviewee has no clue what the lab is even doing and I relay that to my PI.”。●

你被期待带着一个方向到场，而选错方向是有记录在案的被刷理由。 “A top candidate got dropped from second visits because all they talked about was how they wanted to develop their tool for a question that I didn’t think was exceptionally interesting and that didn’t fit into my program at all.”。有竞争力的课题组常常在通话之后要一份一页的书面提案。●

决定背后没有流程。 一位正教授：“I will strategically neglect to request references as I won’t want to be talked-out of a candidate.”。很多候选人根本收不到任何答复。●

关于担保，最有用的一个事实是大多数 PI 根本不知道这套东西怎么运作。 有一份记述说得很直接：大多数课题组长不了解签证机制，所以你必须早早讲明白。大学也倾向于推更便宜的交流访问学者路径，而不是工作签证，因为它更便宜，并且覆盖整个合同期。●

赛道 D. 大型科技公司

所有标准建议都是为这条赛道写的，而它是你投递里占比最小的一部分。

筛选轮是结构化的、事务性的，掌管它的人无法评估你的工作。 招聘人员自己就这么说，而且两个方向都说。一位：“I mean you can’t for the most part. If you were qualified enough you’d be an engineer.”。紧接着另一位：“Bingo! I’m not a mechanic… but I can still screen those candidates.”。来自第三位的调和版本：“Recruiter screen: can the candidate talk the talk… Hiring Manager evaluation: can they walk the walk”。●

这套评估噪声很大，做评估的人自己知道。 一项针对七十六位技术招聘人员的研究发现，简历阅读时长的中位数是三十一秒，准确率在百分之五十五上下，而两位招聘人员评判同一个候选人时的分歧程度，和他们评判两个不同的人时差不多。给出的刷人理由，也和真实理由不一致。◆

薪档是刚性的，这在向下的方向上保护你。 一位招聘人员的说法是：如果区间是十五万到十七万五千，而你说十一万，给出的最低数仍然是十五万。另有一位候选人独立报告说，他被告知自己报的数字太低。这一条不能迁移到赛道 A。●

这对准备意味着什么

如果你只准备一套脚本，它会被调成赛道 D 的样子，而在你大部分投递真正所在的另外三条赛道里表现不佳。

实际的说法是，你需要三种开场，不是一种。在赛道 A 和赛道 C，用研究实质开场，因为电话那头的人能评估它，也期待它。在赛道 B，用一个结构化的行为题答案开场，因为电话那头的人是在拿一套胜任力框架给你打分，评估不了科学。在赛道 D，用一段紧凑的、对事务性条件友好的概述开场，把实质留到下一轮。◇

Q01. “Tell me about yourself.” / “Please walk me through your CV.”

证据强度： ●

他们为什么问： 这是默认的开场题，也是成本最低的一次测试：你能不能把自己的经历压缩成一个非专业听众能跟上的东西。一位招聘人员直白地说，能抓住她注意力的是清晰：“If someone can explain what they did, why it mattered, and how it moved the needle for their team, that person immediately stands out.” 也有几位招聘人员把它当成不计分的破冰环节。

他们在听什么： - 你现在在哪里工作、实际做什么，简短地说，然后是与这个岗位相关的既往经历。一位招聘人员把想听的内容说得很死：“I just want to know 1) where you currently work and what it is you do (briefly!) and 2) any previous relevant & meaningful experience related to the job you are interviewing for.” - 你怎么讲，比你讲了什么更重要。“I listen to them here but I’m more focused on HOW they tell me.” - 你是把自己对到这个岗位上，还是在背一份编年史。

常见死法： 被提到最多的失败是啰嗦。一位候选人把自己的答案草稿贴出来请招聘人员打分，得到的回应是 “90% of your answer did not answer the question…” 和 “I think that’s way too long, and it sounds like chat GPT wrote your response lol.”。一位应届毕业生在电话里当场被拒，因为她 “couldn’t figure out what she wanted to hear”，对方直接告诉她没有准备。反方向的失败同样存在：一位候选人手里 “a list of potential answers ready to go but ended up stuttering through my educational and work background”，那通电话只有四分钟。还有一位准备过头的候选人发现面试官往另一个方向走了：“I prepared so hard that when it wasn’t the exact thing I expected I let my guard down.”

证据内部的冲突： 长度。同一个帖子里的招聘人员分别说 30 秒以内、2 分钟、以及 “3-4 mins is good.”。一位研究岗候选人说自己准备了 “a 30 second story about myself.”。另一批招聘人员则干脆否定这道题：“It’s a terrible question, honestly. I’ve been in talent acquisition for two decades and have never asked this question.” 这个分歧不随赛道走，也不随年份走，它随具体是哪位招聘人员走。把它当作未解决的问题：准备一个 45 秒讲得完的版本，如果对方保持沉默，能扩展到两分钟。

What to say: Three moves. Where you are now and what you build; one line of evidence with a number in it; why this role is the next step. Then stop and let them steer.

“I’m finishing a PhD in [your field], where I build [the kind of method you build] for [the kind of problem it is aimed at]. The work I’d point to first is [the project you would lead with], which [what it changed] by [the number that makes it concrete], and [who uses it now, or what it is used for]. I’m looking for a role where that kind of method work sits close to [the thing this employer has that your current setting does not], which is why this posting caught my attention.”

按赛道： 在研究机构或实验室里，一位被前沿实验室录用的候选人说，同一个位置上出现的是 “Please walk me through your CV”，并补了一个实用习惯：先确认他们是否真的读过你的简历，因为经常没读；如果没读，就按时间顺序把每一站走一遍，再加上你要给的那层解读。在实验室或博后的对话里，这个开场经常被整个跳过，技术追问立刻开始。

Q02. “What made you apply to our company?” / “What do you know about us?”

证据强度： ●

他们为什么问： 为了把选定了这家公司的人，和海投两百个岗位的人区分开。它是构成标准筛选轮的三道问题之一，另外两道是你为什么离开和薪酬。一位候选人在同一个流程里被问了四次 “Why us?”：筛选轮、用人经理面试、技术电话面试，以及现场连面里又问了一遍。

他们在听什么： - 你做过职位描述给不了你的功课。一位招聘人员的说法比较温和：“I don’t expect you to know anything, just be honest and I’ll give you a quick pitch. If you do know, how do you know about us.” - 在小公司，看你能不能用一句话正确说出这家公司到底在做什么。 - 你的理由是不是针对他们的，而不是笼统的恭维。

常见死法： 在初创公司，不了解产品是当场致命的。一位候选人在与创始人通话后写道：“I definitely did not have a deep enough understanding of what exactly the start-up does (which he could tell right away).” 一位在多轮尝试中唯一通过 HR 轮的候选人把功劳完全归给公司调研：说别的候选人不会说的东西，并且表现出你真的了解这家公司。在博后赛道，失败的形态是什么都没读：一位 PI 说自己最近面试了 “a bunch of people recently who had not bothered to read a single paper from our group or even knew what the current state of the art approaches were in our field.”

What to say: Name a specific thing they published, shipped, or announced; connect it to a specific thing you have done; state the open question you want to work on there. Two sentences is enough.

“I read [the specific paper, release, or announcement of theirs] [when you read it] and [what you actually did with it, for example reimplemented part of the evaluation for your own project], so I know where the failure cases are. The gap you flagged in the discussion, [name it in four or five words], is exactly what [the line of your own work that aims at it] has been aimed at, and I would rather work on it with [what they have and you do not, for example experimental feedback] than without.”

按赛道： 对初创公司，创始人或用人经理还想看到你说得出市场：一位创始人自己的筛选问题就是 “Who are our five biggest competitors?”。对实验室或国家实验室，问法会变成为什么是这个实验室，以及你觉得自己会影响这个机构的哪一部分使命，读两三篇他们的论文是明说的下限。对大药企，招聘人员是在打勾，一个笼统但自洽的回答就能过。

Q03. “Why are you leaving your current job?” / “What has you looking for a new opportunity?” / “What are you looking for in your next role?”

证据强度： ●

他们为什么问： 这是一道稳定性与态度的探针，招聘人员还用它给你分流。一位招聘人员说：“I always ask them what about their current job is making them consider other options. It usually gives good insight on what is important to them career-wise and what work environment would be a better fit.” 同一个位置常常变成“你考虑新机会时的优先级是什么”，有位招聘人员刻意把它设计成让候选人说掉接下来 90% 的时间。

他们在听什么： - 朝着某个东西去，而不是从某个东西逃。“Employers generally want the perception that you are running toward their job, not away from a different job.” - 在被邀请抱怨时表现出的职业克制。“There’s a difference between talking shit about your company and telling the truth.” - 一个不只是钱的理由，除非你被严重低薪。

常见死法： 坦率是有记录在案的代价的。一位候选人告诉招聘人员他的雇主对这份工作做了不实描述，之后才知道 “my current employer is a client of her company.”。另一位指出了他认为杀死那通电话的那句话：他说自己在遗留技术栈上 “taken one for the team”，结论是 “it makes more sense to be selfish and protect yourself”，他后来形容这是好的职业建议，但不是管理层想听的话。一位在实习生 HR 轮的候选人如实回答了此前一段实习为何只做了两个月，在总监轮进行顺利之后被 HR 拒掉；她自己给帖子起的标题是，她因为太诚实而被刷。

证据内部的冲突： 该诚实到什么程度。多位招聘人员主张诚实，尤其是关于裁员：“Being laid off is probably the most well-received reason for a employment gap.”，以及 “if comes off better if someone just owns it and is honest. Otherwise it reads like you are trying to hide it.”。与之相对，一位招聘人员说 “It doesn’t matter. You control the narrative… Just give them a plausible reason and move on. It’s not a lie detector test.”，还有人在一则创始人轮被拒的帖子下建议 “giving more cheesy answers next time because honesty is your enemy with HR style questions.”。这个分歧不随赛道走。它随你诚实的对象走：裁员和重组这类事，诚实会加分；对人和对雇主的批评，不会。

What to say: One neutral sentence about the current situation that includes something positive, then the forward reason, then the link to this role. Do not diagnose your employer.

“[One neutral and true sentence about where things stand now, including the part that is going well, for example that the last project is finished and this is a clean point to move]. What I want next is [the forward reason, stated as a property of the work rather than a complaint about anyone], and that is not something [your current setting] is set up to do. That is the part of your posting I want to talk about.”

Q04. “Why do you want to leave academia?” / “Why industry?” / “Are you sure you want this job?”

证据强度： ●

他们为什么问： 他们测的是你的决心，以及你是否理解工业界的工作是什么样，而不是你的动机本身。一位面试过几十位博士的用人经理说：“It’s not meant to be hard or a trap. It’s about getting a sense of where you’re at mentally. You’re fine as long as you don’t come across as only applying out of desperation because you didn’t get a tenure track position. Nobody wants to hire someone who doesn’t want to be there.” 一位候选人在博客里直说，你应该 “think clearly about why you wanna leave (every recruiter and hiring manager will ask you anyways).”

他们在听什么： - 一个计划，而不是一场逃离。一位标注为工业界博士的评论者说：“It is not a bad thing to come from Academia, they just want to see if you have a mid-term, long-term plan in mind for your career.” - 你是不是还在学术求职市场上。一位用人经理：“We often directly ask if they are also applying to academic/post-doc jobs, and why/why not.” - 你谈工业界时的语气。一位面试官点出的最常见自毁形态，是那种把自己在学术界的追求说得高尚无比、把工业界说成逐利的候选人，原话是 “high and mighty about their noble pursuits in academia”。

常见死法： 两种。第一种，答案全在讲学术界如何崩坏。一位候选人的逐字脚本 “Academia was something I was invested in pursuing, but due to how things are it’s simply not a viable career path for me anymore.”，立刻招来另一位候选人的反对：“instead of focusing on what OP wants to do and how they’d be able to contribute to the job they want, it’s more about how academia is all bad and OP just need a job.” 第二种，没有真正消化不同的工作模式。一位评论者认为，如果你 “not thought about what it means to live on administrative time instead of academic time”，也没有为那些冲着学者刻板印象来的问题做准备，那这就是你到了第三轮还在被质疑决心的原因。注意这个疑问会在每一轮被重问，不会在筛选轮就了结：一群转行的前学术人报告说，在终轮面试时被问 “So, you’ve got this PhD, you’ve spent some time teaching, are you sure you want this job?”

证据内部的冲突： 钱能不能说。一位候选人在真实的生物技术面试里这样回答：“I want to make more money than I’m making right now. And I also want my science to have a tangible product at the end of the day.”，这是那个帖子里票数最高的答案。与之相对，一位给资深学术人做顾问的人说，答案应该是 “a compelling scientific journey and impact you hope to make rather than just about money. One topic is viewed more positively than the other.”。把两边都读进去，最稳的解释是：钱可以出现，但不能是承重的那个理由。

What to say: State the scientific reason, name the concrete thing industry has that you need, and close the academic market question before they ask it.

“I want the work to end in something that gets used. In my field the bottleneck is not [the thing your field produces by default], it is [the missing ingredient that decides which method is actually right], and that loop exists here and not in [your current setting]. [Then close the academic market question in whatever form is true for you: which of faculty positions, postdocs, or neither you are still applying to, when you decided, and what you did differently because of it.]”

按赛道： 在制药赛道，问这道题的招聘人员不会评估科学内容，所以要先翻译：“sit down and identify the skills you have acquired and translate them into industry terms/needs.”。在实验室和博后赛道，这道题会反过来，变成一道职业规划题；见 Q07。

Q05. “Walk me through a project you’re proud of.” / “Tell me about your research.”

证据强度： ●

他们为什么问： 研究岗的筛选轮实际上在这里定胜负，判据是推理过程，不是结果。一位在大药企做过初筛和现场面试的科学家说得很直接：“make sure to demonstrate at least one good example of problem solving or your thought process when talking about your research. I saw a lot of people just listing out results without discussing how they designed or iterated on their experiments. We can read your resume, it is really hard to evaluate someone when they don’t do this.”

他们在听什么： - 设计与迭代：为什么做这个实验，你预期什么，什么让你意外，你改了什么。 - 贡献归属的澄清，这是一条明说的判据。同一位面试官：“clearly state what your contributions to research projects were and what was done by your coworkers. Sometimes we can make a good guess at it by inference but it will only help you to remove that guesswork from our evaluation. If you must be modest, the way to go is to clearly indicate everything that wasn’t your contribution so that by process of elimination we know what you did.” - 深度优先于广度。一位做过研究岗求职的机器人学博士说：“it is much stronger to present one thing well and have your interviewers come away with the belief that you really understand it.” - 你到底能不能谈自己署名的那些论文。一位候选人对某大型实验室招聘筛选轮的总结是，那是一个机会，让你展示 “you actually know and can talk about the papers you are an author of.”

常见死法： 论文保护不了你。一位面试官：“Having a bunch of ICML/NeurIPS/ICLR papers is 99% due to your lab/advisor… I’ve interviewed people with very strong resumes on paper that were utterly clueless about fundamental concepts in ML and language modeling.” 一位手握多个顶级实验室 offer 的候选人的结论是 “technical skills and knowledge are evaluated much more than research experience, though the latter probably gets you the interview in the first place.”。在实验室赛道，失败来得更快：一位刚面试完三十位博后的 PI 说，如果候选人声称精通本组的核心方法却答不上来，“If during this technical portion of the interview it becomes apparent a candidate has lied about the expertise I end the interview early.”

What to say: Problem, why the obvious approach fails, what you did, what you personally did versus the collaborators, what it cost, what you would do differently. Volunteer the contribution split without being asked.

“[Who needed what, and the reason the standard approach could not deliver it]. The obvious fix, [the standard approach], was not affordable, so I tried [what you built instead]. My first version [the specific way it failed], which told me [what that failure implied about the setup], so I [the change you made in response]. To be clear about the split: [what a collaborator contributed] was theirs, [what came from someone else’s earlier work] came from a labmate, and [the parts that were yours] were mine. If I did it again I would have [the thing you would reorder, drop, or do first].”

按赛道： 大型科技公司和大型实验室在招聘人员这一环问这道题，只是浅层核实；深挖的版本来自用人经理。小型研究机构可能把它做成一整轮，研究报告最早可能在第二轮就出现。大药企的筛选轮经常把比例反过来，改问情景题；见 Q08。

Q06. “You haven’t worked in this area before. Why are you applying here?”

证据强度： ●

他们为什么问： 他们想知道你是一个能挪动的科学家，还是一个绑死在单一课题上的专才。这是从物理、化学或结构生物学转向机器学习的候选人最担心的一道题，而证据比传闻乐观。

他们在听什么： - 可迁移的技术，不是你原来的课题。一位工业界面试官点出了会被拒的模式：那些 “talk about their pathway/protein of interest rather than the techniques they gained during their careers, and how these can be applied in the position.” 的候选人。 - 在新领域里拿得出的证据。一位转入 AI 安全方向的候选人说，他那些多语言方向的论文，包括最佳论文奖，对结果毫无影响，他称这件事 “liberating, because that means that you can always pivot to a new field… if you demonstrate sufficient expertise in that field and the team wants you.” - 看得懂的产出物。一位核物理博士：“A project using diffusion models can often get you noticed whereas a generic classification task may not.”

常见死法： 以旧领域开场，以及投到那些读不懂你背景的地方。同一位物理博士发现，对物理背景的开放度集中在三个地方：自动驾驶、面向科学的机器学习，以及语言模型初创公司。在别处，转行要从零开始论证。

What to say: Open by converting your background into capabilities the role needs, in one breath, before anyone asks. A candidate hired as a research engineer at a frontier lab used this almost verbatim:

“Hi, I’m [your name], a Machine Learning engineer. You’ve probably read my CV, so I will just briefly summarise what makes me uniquely qualified for this role. My background spans Physics, Sociology, and systems-level software engineering. As such, I can work with quantitative data at scale, have experience with messy human-produced data, and can build the software systems around that to train and serve large models.”

中文大意：我是某某，机器学习工程师。你大概已经读过我的简历，所以我只简要说明为什么我特别适合这个岗位。我的背景横跨物理、社会学和系统层面的软件工程。因此我能处理大规模定量数据，有处理人产生的脏数据的经验，并且能围绕这些搭建软件系统，用来训练和部署大模型。

Your version: name three capabilities, each with a concrete anchor, and end on the one the role is actually about.

“I’ve spent [how long] doing [what your background actually is], which means three things for this role: I can [the first capability, stated as something you can do rather than something you studied], I’ve [the second capability, with the property of the data or the problem that made it hard], and I’ve [the third capability, the engineering one, with the number that shows its scale]. The last [span of time] of that has been [the capability closest to this job] specifically, which is the part of your posting I want to work on.”

证据内部的冲突： 转行的代价有多大。一位在小型研究公司的候选人发现 “most companies and teams did not care too much that I hadn’t previously worked directly on or published in their area of interest.”，一位招聘方的评论者走得更远：“people with deep domain knowledge in something physical that has leveraged ML as a tool are way more interesting to hire than someone buried in code and gradients their entire career.”。与之相对，工业界研究者中一场传播很广的讨论认为，真正的研究岗要求你已经交付过最前沿的结果，并估计全职投入三年才能转进去。这个分歧随岗位走：研究工程和应用科学岗吸收得了转行，前沿实验室里叫做研究科学家的岗位吸收不了，至少不会很快。

Q07. “How does this job fit into your larger career goals?” / “Where do you want to be in five years?”

证据强度： ●

他们为什么问： 在学术赛道，这是整场对话里承重的那道题，而且它要的不是一份文件。一位 PI：“When I say 5 year plan, I don’t mean a form document. I mean that you, a mature, ambitious, and intelligent adult have a clear, and articulatable goals for your future and that you can convince me that spending 5-10 % of your working adult life in my group is going to benefit those goals.” 另一位 PI 说明了这道题测的是什么：“What I want to know is people are going to take active control of their career path… I want people who aren’t passive about their futures.”

他们在听什么： - 一个由你自己把握的方向，以及这个职位服务于它的理由。 - 与他们实际能提供的东西是否对得上。一位招聘人员的说法很简单，职业目标 “should match the role at least roughly.” - 在国家实验室，看的是具体的价值取向：“knowing your interests in starting projects, developing proposal writing skills, facilitating collaborations, all those speak towards the values of the lab and what they want in a new hire.”

常见死法： 答不出来，而且从博后嘴里说出来比从学生嘴里说出来糟糕得多。同一位 PI：“a graduate student not having goals for grad school is a red flag, but a postdoc without them is a fire alarm.” 第二种失败是没看清对面坐的是谁：“Don’t tell me you want to work with a senior PI cause you did your PhD in the lab of assistant professor when i just started my lab.”

What to say: Name the capability you are trying to build, name what only this place can give you toward it, and be honest that the endpoint may move.

“In five years I want to be [the role you are actually aiming at, described as what you would own rather than as a title]. What I still lack is [the capability that is missing, named as a judgment rather than a skill], [the concrete decision that capability governs]. This group is one of the few places where I would sit close enough to that decision to learn it. Whether that ends in [one plausible endpoint] or [the other], I would rather decide after [the span of time] of doing it.”

按赛道： 在实验室、国家实验室和博后赛道，这是一道第一通电话就问的题，而且常常就是那道题。在大药企和大型科技公司赛道，它以招聘人员打勾的简短形式出现。在初创公司的创始人轮里，它通常根本不出现，取而代之的是 Q15 里那些关于风险和适应性的问题。

Q08. “Tell me about a time when…” (behavioral and competency questions)

证据强度： ●

他们为什么问： 在大药企赛道，第一轮很大程度上或者完全是行为面，而且按公开的公司价值观打分。某大药企自己的招聘页面用书面文字告诉候选人用 Situation Task Action Result 方法作答。另一家把自己的胜任力问题逐字公开，包括 “Tell me about a time when you encouraged the sharing of different viewpoints and created a space for others to speak up. What actions did you take to encourage others to voice their disagreements?” 和 “Describe a time when you held others accountable for their commitments. How did you communicate and bring clarity to the team? What, if anything, would you do differently in the future?”。它还写明，面试锚定在它的价值观上，技术问题是可选的附加项。

他们在听什么： - 在制药行业，团队维度高于一切。一位员工点明了这一点：“How have you addressed problems? Have you blazed the trail completely on your own? Or have you sought perspective from key stakeholders and assembled a cross-functional team to establish a well-informed path forward? Big Pharma wants to hear you have done and are willing to do the latter.” - 推理过程，不是简历。一位面试官：“We can read your resume. We want to see how you think and reason. Always plan for situational. Things like: difficult decisions, delivering bad news, showing leadership, unexpected outcomes.” - 在追问下依然具体。一位在大型科技公司做电话筛选的面试官说，她见到的拒绝多半是 STAR 故事讲得不好，或者追问时的回答不够具体；她举的例子是一位候选人引用了一个数字，却说不出这个数字从哪来。

常见死法： 博士候选人对这一轮准备不足，因为它看起来简单。一位手握顶级实验室 offer 的候选人说：“I failed my first behavioral interview because I went into it thinking I’m obviously well-‘behaved,’ and came up blank on excruciatingly simple questions.”，最后面试官说 “You didn’t answer the question.”。第二种失败是手里没有能对上所考价值观的故事。一位质量岗的从业者：“I got HR behavioral questions which were really tough to answer because I don’t generally show courage or heroism at work. I’m in quality… if I’m doing either of those things, the company is doing something wrong.”

What to say: Build six stories before the call, each usable for three prompts, each with a number and a “what I would do differently” ending, because the published probes ask for exactly that. Situation in one sentence, task in one, action in three or four with the decisions you personally made, result with a number, then the reflection.

“[The situation in one sentence, including what had gone wrong between the people involved]. [The first thing you did, and the fact that you did it before proposing anything]. What came out was [the real cause, which is usually not the technical one you assumed], so I [the change you made] and [what you did with it]. [The result, with a number on both sides of it]. What I would do differently is [the ordering mistake, stated as a sequence rather than as a character flaw].”

证据内部的冲突： 这个形式到底筛出了什么真东西。一位在大药企的候选人：“STAR/situational interviews is how companies end up full of good talkers but not many great do-ers. If interviews favor bullshitters and eliminate people who are too truthful or stutter, then companies shouldn’t be pikachu surprised when the chain of command spins bad news as good all the way up.” 同一批帖子里的面试官则坚持，行为面的权重是有意为之而且是对的。两边都同意它在大药企是普遍存在的，所以这场争论不改变你要准备什么。

按赛道： 大药企和大型科技公司：预料到它，准备它，用 STAR，因为至少有一家公司白纸黑字要求这么答。小型研究机构：基本没有，取而代之的是 Q09。国家实验室：有，但范围很窄，常以小组面的形式出现，问题是写好的脚本，围绕论文写作、技术分歧、指导他人，其中一家实验室问的大多是安全。

Q09. “Here’s a problem we’re working on. How would you approach it?”

证据强度： ●

他们为什么问： 小型研究机构用一道没有既定答案的问题取代行为面，然后看你怎么做。一家研究型初创公司这样描述自己的流程：“giving them a real problem we’ve recently faced and seeing how they would reason through it.” 另一家的候选人被要求准备一个 30 分钟的报告，主题是 “on a problem they had that I had no domain expertise in.”。在一家 AI for science 公司，20 分钟的第一通电话之后直接给了书面作业：了解我们的平台，说明你会怎么用它做一种免疫疾病的药物筛选。

他们在听什么： - 迭代速度，这是明说的。一位在研究型初创公司面试博士的工程师：“I would rather hire someone who comes up with a sub-par solution quickly and then iterates on it, than someone who comes up with the perfect theoretical solution, but no code.” 他给博士候选人的唯一一条指示是 “Just come up with something.” - 澄清，这是被允许也是被期待的。同一位面试官：“It’s always important to have a discussion … to understand the problem.” 问清楚约束是什么、有哪些数据、什么算成功，这是答案的一部分，不是答题前的拖延。 - 在自己知识边界上的诚实。一位研究岗候选人：“Be honest about what you don’t know… this came across much better than trying to bluff through a topic I wasn’t confident about.”

常见死法： 一边找正确答案一边沉默。还有基本的工具能力：在面试过几十位博士之后，一位面试官说最常见的淘汰原因是 “Most PhDs we’ve encountered have very basic programming skills.”

证据内部的冲突： 这道练习到底测了什么。一种解读是科学判断力。另一种来自对同一个说法的批评者：“The startup risk tolerance is extremely high, thus it rewards quick and dirty over well thought out and solid.”，第三种则质疑这个推断本身：“Notice the assumption here, that their 20-minute task is a perfectly reliable indicator of how the interviewee would behave in general.” 一位相关的评论者把它描述成一种结构性错配：博士的价值主张是一份复杂、深思熟虑的学位论文，而这道练习要的正好相反。坐在候选人的位置上你解决不了这个问题。按他们的方式打，同时留意这件事告诉了你什么关于这份工作的信息。

What to say: Out loud, in this order: restate the problem, ask two clarifying questions, state your assumptions, propose the crude version, name how you would know it is wrong, then name the better version you would build if the crude one survives.

“Let me make sure I have it: you want to prioritize compounds for a screen, and the constraint is assay throughput rather than compute. Two questions: how many can you actually run per week, and do you have any negatives, or only the hits? Assuming a few hundred per week and hits only, the crude version I would build this week is a similarity baseline over your existing hits, no learning at all, so we have a number to beat. The way I would know it is wrong is if it retrieves only analogs of what you already found. If it survives that, the version I’d actually want is a model trained on the assay readout directly, and the first thing I’d check is whether the readout is stable enough across plates to train on.”

Q10. “It doesn’t look like this is on your resume, but I want to be sure.” (the verification probe)

证据强度： ●

他们为什么问： 这一轮有很大一部分存在的意义就是确认简历是真的。招聘人员反复这样说：“I can tell if their resume is inflated or accurate, if they can communicate clearly, if they can speak to their past projects.” 一位主张自己来打第一通电话的用人经理：“Cv is usually 90% buzzwords and things the person didn’t actually contribute to. Why not figure this out on the first call with him, asking a couple experience related questions instead of letting hr do this.”

他们在听什么： - 你能不能比简历上那一条再往下走一层。“Sometimes people claim to do things on their resume but when probed, they cannot really speak to them or give vague answers and cannot give me a specific example of when they actually did the thing.” - 经得起追问的量化说法：“if you’re telling me you did something ‘well’, I want to know what you specifically did and how you’re defining doing it ‘well’.” - 校准过的自我评估。一位招聘专员：“You’ll get much further being honest with experience level than when you tell me you’re experienced with a specific system, and your follow up statement demonstrates you have little to no familiarity.”

常见死法： 一位候选人把自己通过的那次筛选轮形容成 “very much an ‘are you BSing or is this experience actually genuine?’ round,”，其中包括一个咄咄逼人的问题，问他怎么做到一边全职工作一边读完硕士。另一位通过了筛选轮里整个行为面的部分，却在一小时内因为一条工具声明被拒：“my CV says intermediate Excel but, I am not and the role is looking for an intermediate user.”。拒信里就写了这个理由。

What to say: Answer at the level of a specific instance, and mark the boundary of your experience without apologizing for it.

“I’ve used it in production on one project, [the project], where I [the specific parts you built with it] myself. I have not [the boundary of your experience with it, named plainly], so if that’s what the role needs on day one, I’d be learning that part.”

Before the call, go down your own resume line by line and, for anything you cannot expand into ninety seconds of specifics, either prepare it or remove it.

Q11. “When would you be able to start?” / “What’s your availability?”

证据强度： ●

他们为什么问： 这是极少数普遍会问的问题之一。一位招聘人员：“The only universal questions I ask all candidates are … what’s their availability timeline.” 另一位把可到岗时间列为申请表上两道经典硬筛之一，另一道是工作许可。招聘人员交接给用人经理的备注里，至少包含候选人为什么在看机会、他经历的摘要，以及 “any timeline constraints for the candidate to interview or start.”

他们在听什么： - 一个真实的日期，以及它是否匹配这个岗位编制的紧迫程度。 - 离职通知期，以及任何会影响后续各轮排期的因素。

常见死法： 含糊本身不直接让你付出代价，但它拖慢流程，而岗位编制是有时钟的。一位用人经理描述自己一周筛一百份申请，并说一旦有四五位好候选人进了第二轮，“I’m not really revisiting the incoming resumes again.”。在博后和国家实验室赛道，错配是反过来的：一位 PI 指出，有兴趣和有经费是两件独立的事，而实验室的时间线以月计，所以你报出的开始日期换来的可能是长时间的沉默，那不是拒绝。

What to say: Give a date and one constraint, then hand the timeline question back.

“I defend in [the month you defend] and could start any time from [the earliest date you would actually accept]. The only fixed constraint is [the one commitment already fixed in your calendar, with its dates]. What does your timeline look like on your side, and how many rounds are there after this one?”

按赛道： 大药企以月为单位推进，日历慢是常态；多位候选人报告，光是从 HR 筛选轮到用人经理就要十到十四天。初创公司可能一周之内从第一通电话走到 offer。一位国家实验室的候选人从面试到收到消息隔了两个月，另一位十一天就拿到 offer，只因为该部门主管知道有一个竞争性的截止日期。要问对方的时间线，因为各赛道之间的答案差着一个数量级。

Q12. “Are you interviewing anywhere else? Where are you in those processes?”

证据强度： ●

他们为什么问： 招聘人员给出几种彼此不同的动机，它们不是同一个动机。管道管理：“If you are in final stages with 3 other companies, I’ll never be able to move you through our process fast enough to compete.” 在他们自己这边争取筹码：“if I like you, and you’re approaching finals somewhere, I’d like some leverage to use with my HMs to make sure they get you in ASAP.” 以及对候选人投递量的检查：“1 or 2 other (early) processes are fine. But some people have 12 things going on for all kinds of places, which is a red flag.”

他们在听什么： - 你是一个真实的候选对象，还是一张彩票。 - 他们的流程是否快到足以赢下你。 - 一致性，因为之后会被核对。一位候选人警告：“Treat recruiter calls as if you are talking to a permanent recording machine.”，并说招聘人员会交叉核对你先前关于竞争性 offer 的说法。

常见死法： 把在跑的流程数量说得虚高，以及对猎头招聘人员报出公司名。

证据内部的冲突： 回答会不会伤到你。企业内部的招聘人员说不会：“We always expect candidates to be interviewing other places. Answering that question shouldn’t deter your chances in any way.” 一位猎头方的招聘人员在公司名这件事上说法相反：“most recruiters will then call A and B company and get those jobs on and put their candidates in… you are diluting your application everytime you tell someone where you are going to.” 这个分歧随提问者的身份走。企业内部招聘人员，答到“进行到哪一阶段”这个层级。猎头招聘人员，答到阶段这个层级，公司名不说。

What to say: Stage, not names, plus a live deadline if you have one, because a real deadline is the one thing that compresses these timelines.

“Yes, [how many other processes and what stage each is at], and [the nearest real decision date, if there is one]. That’s not pressure, it’s just the calendar I’m working with. If your process runs longer than that, it would help me to know now.”

按赛道： 在博后赛道同样会问这道题，而学术那边给的建议更强硬：“Make sure not to hide if you have applied elsewhere; scientific world is small, and people talk.” PI 的版本通常还会加上 “what criteria would you use to decide between competing offers”，而被打分的正是这些判断标准。

Q13. “What’s your biggest weakness?” / “What’s the most recent piece of critical feedback you received?” / “How would your last boss rate you?”

证据强度： ●

他们为什么问： 看你对自己有没有一个可用的模型，以及你能不能在谈批评时既不崩溃也不推开。招聘人员的标准题库里都留着这道题的某个版本：“What are you not good at or not interested in doing professionally? (push for at least 2-3 examples and why).” 一位招聘人员更锋利的变体：“Who were your last three bosses, and how will they each rate your performance on a 0-10 scale when we talk to them? … it’s not that you’ll actually ask, it’s that you get a sense of how the person responds to management.”

他们在听什么： - 具体、近期、真实。一位创始人：“Very specifically, what’s the most recent piece of critical feedback that you’ve gotten?… I’ve found that throwing in the ‘references’ comment is important because it tends to bring out more honest responses. I’m literally looking for them to get into the nitty-gritty of the when and the how, not fluffy or abstract responses.” - 你能不能在不带轻蔑的情况下承受别人对你的负面看法。一位创始人会问 “If I were to go and speak to people who don’t think very highly of you, what would they say?”，然后观察你是回避、辩解，还是否认有人不喜欢你。 - 在某大型科技公司的实习生 HR 轮里，这道题以压缩形式出现：用两三个关键词概括你的不足，再给出三个你的合作者会用来形容你的关键词。

常见死法： 抽象，以及变相自夸。上面那位创始人的话针对的正是 “fluffy or abstract responses”，而提到“背景调查”这个框架之所以存在，就是因为候选人否则会给出一个安全答案。

What to say: Name a real limitation, give the incident, give the mechanism you now use, and give the current status honestly.

“The most recent piece of critical feedback I got was from [who gave it] after [the occasion]: [the feedback in their words, not softened, and what it means in practice]. It cost me [the concrete consequence]. Since then [the mechanism you actually use now], and [the check you built in so it does not depend on your memory]. It is better, not fixed; [the condition under which you still fall back into it].”

Q14. “How do you work with AI?”

证据强度： ◆（来源数量少，且都来自同一个平台；证据薄弱）

他们为什么问： 一位在北美找工作的数据科学家说，越来越多的招聘人员现在喜欢问 “How do you work with AI?”，它是招聘人员的常规问题，不是技术问题。在某大型科技公司的实习生 HR 轮里，这道题是一个正式计分项：你实际用哪些智能体和工具，你关注哪些 AI 信息和模型，要给出具体细节。

他们在听什么： - 具体的工具和具体的用法，不是观点。 - 你的做法与团队的做法是否一致。一位在一家融资充裕的 AI 初创公司面试的候选人问团队自己写多少代码，得到的回答是过去三个月他们 “almost exclusively using claude/codex on their research teams, as it’s allowed them to spend much more time experimenting and ideating, and leaving the execution to the bots.”

常见死法： 毫无准备被抓个正着。同一位数据科学家说，最初几次这道题让她措手不及，她答得没有结构，而且不止一次在招聘人员这一轮被刷掉。

What to say: Two sentences of practice, one boundary.

“Day to day I use [the tools you actually use] for [the parts of your work where the specification is already clear, named concretely], and I keep [the parts you do not hand over] under my own hands, because that is where a wrong answer is expensive and silent. The habit I’ve built is [the one rule you actually follow, stated as an ordering].”

Q15. “Why do you want to work at a small company?” / “How do you feel about ambiguity?”

证据强度： ●

他们为什么问： 小公司在筛的是离职风险，以及你是否需要他们给不了的结构。一位候选人对初创公司筛选轮真正在测什么的描述：“They are also vetting you to see if you’re just taking the job for a paycheck and will continue interviewing for bigger companies. Big company employees don’t work out as good hires at startups usually.” 创始人们公开说明适应性这条判据：“anyone you hire is likely to be doing a new job in three to six months. Smart and effective people are adaptable.” 以及：“The worst employees for a startup are those that come in and ask for the training manual. At a startup our job is to figure it out together, no one will spoon feed you a process.” 一位 A 轮前的创始人直接问：“What is your appetite for risk? There is a high chance of failure.”

他们在听什么： - 一个选择小公司的理由，而不只是因为这里刚好有个坑。一位候选人说，他在一家大型科技公司工作过之后，反复被问为什么想去小公司，并指出 “saying better opportunity does not fit well.” - 你在没有明确职责范围的情况下做过事的证据。在科学类初创公司，这就是岗位描述本身：“You often create your own job responsibilities as you go.” - 广度。具体在生物技术领域，“startups want someone who is a generalist, not a specialist, with as much experience as possible in a broad number of areas.”

常见死法： 一位候选人用两句话记录了整段交流：对方问他喜不喜欢初创公司，他回答不喜欢，流程到此结束。另一种失败是反向的，值得知道：来自大机构的候选人在初创公司这边会被反向筛掉，一位评论者形容那些大公司出来的人 “can’t stick to a small budget or expedite a timeline.”

What to say: Give the concrete reason, then the evidence, then the honest limit.

“I want to be close enough to the decision to change it. In my group I [the list of things you personally carried, from building the thing through to writing it up and dealing with whoever else was involved], and the part I liked was that nobody handed me the scope. What I’d want to know from you is how the first six months are defined, because I do better with a clear objective than with a clear process.”

按赛道： 这道题属于小型研究机构和初创公司赛道，而且常常由创始人或 CEO 亲自问。大型科技公司和大药企不问。注意反向的问题也存在：在制药赛道，从初创公司回到头部公司，被一位有十八年经验的从业者形容为困难，所以你移动的方向本身就被读成一个信号。

Q16. “What should we be doing differently?” / “I’m not sure you’re the perfect fit. Why do you think you’d be great?”

证据强度： ● 就这个做法本身而言；至于要不要主动提出批评，证据直接自相矛盾

他们为什么问： 在创始人轮里，这是一次人造的反对意见，你的反应就是数据。一位创始人：“This is a great question because it’s an objection. I want to see how they handle objections like they would from a potential client or from a co-worker with a contrarian opinion.” 另一位投资人兼经营者描述了他据以招人的先行指标：“The candidate can see things you don’t see. Even within topics you’re fluent in, they’re able to convince you of new points of view or make you realize you’re missing something.” 一位销售负责人的版本是 “What should our team be doing differently that could yield 10x improvement?”

他们在听什么： - 你到底有没有一个看法。关于那道 10 倍的问题：“Can’t think of any suggestions. This one is a big red flag for me… They should be pretty well-versed in our company and product by then, so it’s often a sign that they haven’t done their homework.” - 建议的野心：“Oftentimes I hear ideas that are a 10% improvement, not 10X.” - 当反对意见冲着你来时，看的是镇定，不是正确。

常见死法，以及证据内部的冲突： 未经询问就主动批评，是有记录在案的代价的。一位通过了技术轮的候选人，主动向一位联合创始人分享了对公司产品战略的看法，随后被拒；评论者的诊断很直接：“where you asked for an opinion in this one? Because if you shared that perspective without being asked to, I see why you were rejected.” 反方规则最强的版本是：“Having been interviewed directly by the founders of the start-up means the company is a) still small enough and/or b) the founders still care very much personally whom they employ. If in such context you are applying as a rank and file software developer the interview is NEVER a good place and time to discuss your thoughts on their business if not being asked.” 这与上面那些把未经提示的批评当作他们正要买的信号的创始人直接冲突。这个分歧随两件事走：你是否被邀请，以及批评落在某个功能上还是落在商业模式本身。在低于制定战略的层级上，否定你正在面试的这家公司的商业模式，就是会失败的那个版本。

What to say: If invited, go one level deeper than a process tweak, and frame it as a question you would want to test rather than a verdict.

“One thing I’d want to test in the first month: [the public claim of theirs you actually read], but [the pattern you noticed that the claim does not account for]. If that holds, the ten percent version is [the incremental fix]. The ten times version is [the structural explanation, stated as a possibility rather than a diagnosis]. I don’t know which it is from the outside, and that’s the first thing I’d want to find out.”

If not invited, hold it, and use Q17 to ask the question instead of stating the conclusion.

Q17. “Do you have any questions for me?”

证据强度： ●

他们为什么问： 一部分是推销，一部分是信号检查。一位用人经理描述自己的第一通电话主要是介绍情况，他说 “the goal with the first call is that they’ll ask the majority of the questions.”。一位招聘人员直言，问题的内容不如“你问了”这件事重要：“recruiters often aren’t evaluating you too much based on the content of your questions, but they do notice the impression you give… Showing that you’ve prepared already puts you ahead of the curve.”

他们在听什么： - 准备程度，以及关于团队、文化和长期方向的问题，而不只是流程事务。 - 在创始人这一层，问题被直接打分：“people who are very interested in joining a company have done their homework. They typically have several detailed questions, and those questions reveal a lot of things those people find important in their job.”

常见死法： 一个问题都没有。还有问了一个被面试官视为越界的问题：一位候选人问第二轮的面试官，他认为团队的管理有哪些地方需要改进，被告知这是入职之后再问的事，如果你被选上的话。评论者诊断的是措辞，不是话题。另外注意这个环节并不保证存在：一位候选人四分钟的筛选轮结束时根本没有它，“They didn’t even ask me if I have any questions.”

证据内部的冲突： 时机。一位招聘人员说什么都不该留着：“The best questions you should be asking in an interview don’t really change depending on the stage… Don’t hold anything back thinking you’ll save it for the next round, because the next round isn’t guaranteed.” 一位在这一行做了三十年的招聘人员对一切涉及弹性工作或福利的问题给出相反的建议：“I always coach that it’s best to wait till they want you. Then you’re in the driver seat.” 这个分歧随问题的类型走，不随阶段走：关于岗位和团队的尽职调查放在前面，雇佣条件放在后面。另一个分歧出现在收尾问题 “do you have any concerns about my fit that I can address?” 上。两位经理说它效果很好，其中一位说 “I thought this was brilliant because I did circle back to some of her gaps.”。一位面试官则说法完全相反：“NEVER ask that question during an interview. It directs the interviewer to deliberately think negative thoughts about you that they will not be able to forget.”

What to say: Two or three questions the person in front of you can actually answer, and one that tells you something real.

“Three things. Why is this role open, is it new headcount or a replacement? What does the rest of the process look like and how many rounds? And in the first sixty days, what is the problem you would want this person to have made progress on?”

Then, depending on lane, one substantive question: for a startup, how the research agenda is set and how much runway is behind it; for a lab, what happens to a project you start if you leave; for big pharma, what the hiring manager’s team is measured on.

按赛道： 招聘人员回答不了研究问题，也不该被问；多位招聘人员明说自己不是那个层级的专家，技术问题属于用人经理那一轮。把它们留着。在小型研究机构赛道，第一通电话可能就是创始人，那种情况下值得花时间的只有那些实质性的问题。

这个问题实际上是干什么用的

招聘人员向你要一个数字的时候，她不是在评估你的判断力。她是在拿这个数字跟一个已经存在的薪档做比对 ●。一位招聘人员这样描述她自己的筛选轮：“My primary focus is to make sure the candidate is ok with pay range, location, schedule.”。一位经历过大量制药行业筛选轮的候选人把这通电话描述为 “3 minutes to discuss your background, 3 minutes to tell you about the job, and 3 minutes to talk about salary.”。另一位把机制说得更直白：“there’s a salary range for each position (it may be broad at your level, but the limits will be strict), and they want to make sure that your expectations are compatible with theirs.”。

从招聘方那一侧看，这个比对明确就是一刀 ●。一家小型生物技术公司的用人经理：“if you aren’t pursued, they will typically screen you out based on compensation being too high or if the recruiter doesn’t believe the personality is a good fit.”。一家大型科技公司的招聘人员：“Recruiters have always rejected based on compensation expectations, even during better times. If you’re out of budget or at the max end of the budget and there are other candidates just as capable and happy with $10-15k less, it makes sense.”。同一个人补充了候选人通常漏掉的那一半：“Comp expectations are considered the whole way thru all the way to offer. I’ve chopped candidates at offer stage due to bad faith or incredibly unreasonable aggression around compensation.”。

有一位招聘人员把容忍度量化了：“If you’re in range but top of band most teams will still interview. If you’re +20-30% of band, it may not be worth wasting time” ◆。

还存在一种和“付不付得起”完全无关的筛选。一位 C 级高管：“We coach all hiring managers to look for the signs that a person will be dissatisfied with their pay as it’s a big giveaway that they won’t stay long in the company. If your desired salary is above our target and especially if your level of experience falls below our target, we know from experience that you’ll be onto a new job within a year. It’s better for us to pass on you” ◆。一位招聘人员用一个具体案例给出了同样的机制：某候选人在 80 到 90 的预算下要价 95 到 100，而其他人要价 75 到 85，结果 “They didn’t get interviewed; the salary demand itself wasn’t the killer […] the problem was that they were simultaneously asking for more (could be done for a great candidate) while offering less than other candidates in the same pool. If you’re asking for more, you need to be demonstrably better” ◆。

公开发言与匿名发言之间有据可查的落差。 这是整个证据集里最尖锐的一处分裂，而且它精确地对应招聘人员是不是在用自己的实名说话 ●。

在实名场合、在公开的职业平台上，招聘人员说这个问题是善意的。“When we ask about your expected salary, it’s not to undervalue you or judge your worth. It’s to ensure that both sides are aligned from the very beginning.”。“So next time a recruiter asks about your salary expectations, see it not as a ‘trap,’ but as a step toward an honest and fair hiring journey.”。一位大型科技公司的前招聘人员在自己的频道上说：“when a recruiter asks you what are your salary expectations it’s not a trick question it’s actually them being genuine”，而且说得更重：“hiring managers if they really like you want to give you as much money as possible cuz they know you’re the perfect fit and they’ll put their neck on the line to um extend their budget.”。

在匿名场合、就同一个话题，招聘人员描述的是筹码和隐瞒 ●。“Salary is also an instant rule in/rule out.”。“Upper executives ask us not to disclose top range unless we need to.”。关于为什么不把区间告诉候选人：“People are greedy and the vast majority will just say they want the top of the range, which makes your job 10x more difficult because now they know the top end… You’re letting the candidate control the process instead of vice versa.”。一位猎头描述自己主动把候选人往下压：“If I know the candidate is earning $90k and wants to jump at the $130k mark, it’s my job to hammer them down back to reality to the $110-115k mark.”。还有一位候选人报告说，有招聘人员直接告诉他，这个数字关系到招聘人员自己的报酬：recruiters “know it’ll give them a leg up in negotiations, and some of them are even paid specifically to get that number (as one recruiter was brazen enough to reveal to me)” ◆。

实名的公开口径连自身都不自洽。那位公开说这个问题是 “genuine” 的招聘人员，在别处的建议是：“the reason why you want to redirect this question is because when you’re in a recruiter phone screen… you actually don’t have enough information to be able to tell the recruiter what your expectations are how would you know” ◆。

这一条怎么用： 把公开口径当成市场宣传，把匿名口径当成实际操作，但不要把匿名口径理解成恶意。两者描述的是同一个动作：你的数字被拿去和一个薪档比对，不匹配就结束通话。每次筛选轮之前对自己说一遍：这是一次比对，不是一次评估。

那个对照实验

有一位候选人做了整个证据集里唯一的一次对照实验，这也是本节最有用的一条 ◆：

“I experimented with salary requirements! I applied for the same job under different names and with very similar resumes (keeping key words the same but making minor changes to the layout). Out of the 20 jobs that required my expected salary, I used the minimum, average, maximum, and above maximum. Most maximum or above-maximum salary applications received a rejection email within one or two days. The minimum salary always resulted in a first-round interview.”

中文大意：他用不同的名字和高度相似的简历投同一个岗位，在 20 个要求填写期望薪资的岗位上分别填了区间下限、中位、上限和高于上限。填上限或高于上限的申请多数在一两天内收到拒信，填下限的每一次都拿到了第一轮面试。

要看清这件事证明了什么、没证明什么。它证明了在投递和过滤阶段，这个数字是一个单调过滤器：报低进得去，报高被快速拒掉，资历完全相同也救不回一个高数字 ◆。它没有证明后续 offer 的任何事，因为根本没走到 offer 阶段。这是一个人、二十个岗位、自述的实验设计、没有报告效应量。把它当作关于过滤器的强证据，同时当作关于 offer 的零证据 ◇。

本节其余内容全部建立在这个区分之上。

照着雇主自己公布的区间报价，被当成要价太高

这是最能击穿标准建议的一种模式，而且反复被佐证 ●。

一位候选人一字不差地报出了公布的区间：“I was asked for my expected salary during an interview. I replied that the salary range in the job posting was exactly what I was looking for. I was rejected the next day by the recruiter because my salary expectations were too high. You cannot make that up. Obviously, they were looking to lowball and had no intention of ever paying the posted range.”。

一位候选人满足了全部必需项和加分项要求，包括一门稀缺语言：“I state it would be on the higher end of the range provided due to meeting not only minimum requirements but also all preferred requirements too. After this the demeanor completely changed and I was essentially rushed off the call… I have just received my rejection email.”。

一位候选人要的是招聘人员亲口说出的真实预算的上限，不是公布区间的上限：招聘信息写的是 76,000 到 116,000，“I ended up saying $95k which isn’t much more than my current salary. They told me that their budget was $90-95k and they were required to put that large range. Anyway, I was very disappointed when I ended up being rejected.”。

一位候选人在招聘人员给出的 100 到 140 之下要了 110：“I said 110k, and the next day I got an email saying the manager thought that was a bit too high, but maybe they could work something out. Then a couple days later, I got a rejection.”。第二位候选人在一次求职里两次遇到同样的形状：“Both said the interview went well, asked what i wanted / confirmed what i initially asked for, then rejected me. The amount is usually not crazy high either. It is actually less than what i make now.”。

还有两个案例，对方给出的理由不可能成立 ◆。一位候选人按区间上限回答，随后被失联（对方不再回复），然后 “I saw that they had reposted the position and this time the lower end was 10k higher than I asked for.”。另一位从相反方向报告了同一类事件：一次面试中 “his mood shifted right after answering his question about my salary expectations which was $25 per hour”，接着被失联，接着同一个岗位以每小时 18 美元重新挂出。

招聘人员从他们那一侧确认了底层规则。“you’re not getting the top of the band unless you’re practically overqualified for the role.”。“If you go for the top of the range and you don’t have the experience for that, you’ll get a lower offer or won’t be considered because you’re too expensive in relation to your experience.”。按该问题下得票最高的回答，最让招聘人员反感的是 “when candidates flip the question and I give a range and they immediately tell me their expectation the top of the range” ●。

这一条怎么用： 在筛选轮里不要报出任何区间的上限，无论这个区间是公布的还是口头说的。如果你要用公布的区间，就把它当成一句“对齐”的表态，而不是对它天花板的主张。一位候选人报告说下面这句话从来没有引起过警觉：“I’m comfortable with the salary range shown in the job ad, and if possible, I’d like to discuss my salary expectation further later in the process.”

回避这个问题的真实代价

回避的成功率高到有人会推荐它，失败起来又惨到无法挽回。两侧都是第一手证据。

回避失败的情形。 最清楚的一例是一通四分钟就结束的电话 ◆。候选人在第三分钟被问到数字，说他需要先了解福利和股权的情况，得到的回应是：“No, just forget about benefits and stock options for now, I just need to know what number you’d be ok with.”。他没有让步。招聘人员：“Woah woah man, you’re not even there yet. We can talk about that later but I NEED to know what is negotiable now.”。然后招聘人员说了句 “Ok, have a good one”，在候选人来得及道别之前挂断。一周后拒信到达。

同样的形状，只是没有火气 ◆：一位候选人把问题抛了回去，“If you could tell me what the available budget is for this position, then I can confirm whether it’s suitable for me or not.”。面试官笑了笑说 “Okay, we can skip this point for now.”。候选人自己的记录：“After that, the whole vibe changed. And just as I expected, of course, no one contacted me again about the third round.”。

在通话之前就索要薪资区间，有它自己的失败方式 ●。“Asked a couple of recruiters for salary bands before the call and got ghosted lol.”。“I asked him the position, tasks and the pay range: He replied with everything, except the pay range. I asked again the pay range. Ghosted.”。从招聘人员那一侧看，这不是偶然：“In my experience if you send an outreach message and what you get back is ‘what’s the comp’, 95% of the time they never meet about the role.”。

回避还经常根本撑不过第一次接触 ●。“I tried to dodge but recruiter said it was mandatory.”。“during the onsite prep call with the recruiter he required me to give some sort of salary range… or I couldn’t go ahead with the onsite interviews.”。“they asked me my salary expectation and I said ‘whatever is the industry standard’ and they kept persistently asking me for a number.”。“Even though I tried to defer this, they got it out.”。招聘人员点名了具体哪些说法会失败：“Don’t tell the recruiter you’re looking for ‘market rate’ or ‘I’m sure you’ll make me a fair offer.’”。

招聘人员还描述了过度回避在观感上的代价 ●。“If you dance around the question (‘I will wait to evaluate the total compensation’), you’re just gonna piss them off.”。“Don’t play games. Tell me exactly the comp you are wanting and I will tell you exactly if it’s within range or not. It’s a stupid game candidates play and is an immediate turnoff.”。一位候选人说出了社区收敛出来的规范：“If you want to deflect the first time the question is asked, fine. But if you have to deflect more than once, you won’t come off like a professional.”。一位招聘人员把这条线画得更精确：“Asking ‘what’s the budget?’ or ‘what’s the range?’ is completely fine. In fact, it’s a good question… But where people get it wrong is they rely on that as a rule. If you keep deflecting and refuse to give any indication, it can come across a bit evasive, especially later in the process.”。

回避奏效的情形。 一位候选人比较了自己的两种做法：“Some of the ones where I did not state my expectation mostly went silent; with the others the recruiter stated the company budget up front. The first group went nowhere, the second group basically all reached the interview stage.” ◆ 注意，这条其实是回避失败的记录。

与之相对的是：“I always respond with, ‘what is your expected range?’ That has always worked” ◆。有人硬顶了公司声称的政策：“they said they must have my previous salary because it is their policy. So I responded that it was my policy not to disclose my salary. They couldn’t argue with that, and I got the offer I was looking for” ◆。“I declined to answer with the role I have now, and it worked out :shrug:. Not everyone is trying to screw you” ◆。以及在一家大型科技公司，“my recruiter asked before interview but did not press when I declined” ◆。

这个分歧对应两个变量 ◇。第一，资历和稀缺性：一位候选人报告说回避话术管用，并补充 “you will see some companies pull out from it, you need to be prepared for that. It’s more acceptable the more senior the position is.”。第二，招聘人员是否已经被要求在安排更多时间之前先拿到一个数字；只要这是硬性要求，任何措辞都撑不过去。

这一条怎么用： 只回避一次，形式是反问他们的区间，然后就给答案。把它当成一个动作说完，而不是两个：“I don’t have a specific number in mind yet. Could you share the range that’s budgeted for this role, and I’ll tell you right away whether it works?” 如果他们回答了，就针对他们的数字回应。如果他们顶回来再问你的数字，立刻给一个区间，不要道歉。不要回避第二次。

报低了会不会把 offer 封顶？

证据分裂得很干净，而且分裂线对应组织类型。这不是一个求平均的问题；两侧在各自的赛道里都是对的。

报低不会把你封顶的地方。 一家大型科技公司的招聘人员：“Sharing comp expectations that are too low will NOT result in a lower offer. Why? Because HR wants to make sure internal equity is fair and the hiring manager doesn’t want you to leave the moment you get a better offer. If the range is 150k - 175k and you tell me your salary expectations are 110k, the MINIMUM I’m going to give you is 150k.”。他自己划定了适用范围：“this is for large tech companies and FAANGs” ◆。

其他招聘人员的佐证 ●：“If you’re too low and we move forward with an offer, most likely you will get more than your ask. Positions have to pay people in a similar ballpark for the same job.”。“If they are looking for the low end of the range but have everything we want - we will pay them the high end.”。以及来自接收端的候选人 ◆：“The few times I gave a lower number the recruiter told me my number was too low and what the market rate is.”。一位在旅游科技公司把自己报低了的候选人：招聘人员的反应是 “Really?…”，然后主动在候选人自己的数字上加了五千美元。

在大型制药公司，薪档下限起到同样的保护作用 ●：“most, if not all, major biotech/pharma companies have a policy to not pay outside of brackets be it above or below.”。“if you low ball the range companies usually have a policy where they have to pay within a certain band even if your original salary request was lower for roles.”。

报低就变成 offer 的地方。 “Well, made a mistake on telling the recruiter salary expectation prior to the offer stage. No surprise, recruiter came back with an offer exactly what I asked for and told me that you got what you wanted and pls sign. I didn’t know what happened on that day, the number i asked for was my current TC” ◆。

一位候选人在对方声称的 60 到 70 预算下报了 65 到 70，拿到的 offer letter 是 60，低于他自己所报区间的下限 ◆。一位在筛选轮里报了数字的候选人，偶然从另一位做同一个岗位的招聘人员那里得知，那个岗位的价码 “10-15K higher than what I quoted the screening lady” ◆。一位泄露了薪资历史的候选人：招聘人员 “refused to put me through for that saying that it was not usual to get more than 10% more than what I am on. The company was willing to pay more” ◆。而且这条长长的后悔尾巴是真实存在的 ●：“The first time I did not dare to ask; the person who got in was at least 5% above me”，以及 “I also asked too low when I joined last year, and I have regretted it all year”。

有两个结构性事实让这个数字变得持久 ●。一位招聘人员：“recruiters keep very detailed notes so we know exactly what salary expectations you shared with us.”。而一位招聘人员交给用人经理的交接表，第一栏就是 “comp expectations”。

这个分歧对应什么： 它对应雇主是否运行一套带内部公平政策的正式薪档，以及流程中是否有人从这个差价里获得报酬 ◇。大型科技公司和大型制药公司有正式薪档和薪酬压缩政策，证据的两侧都同意报低会被拉到下限 ●。小公司、猎头，以及任何“听你报数字的人同时也是写 offer 的人”的流程，恰恰就是数字直接变成 offer 的地方 ●。

这一条怎么用： 在任何赛道里都不要报出一个低于你能接受的数字，因为那层保护是雇主的属性，不是你的属性。在你不确定赛道的时候，就把这个数字当成有约束力的。

阶段之分，这是整节的钥匙

上面几乎所有表面矛盾，只要把两个用同一个问题服务相反目的的阶段分开，就都消解了。

第一阶段，过滤。 招聘人员在拿一个数字比对薪档，决定要不要再花时间。招聘人员明确这样说 ●：“Recruiters are not negotiating when initially discussing salary expectations, we just wanna make sure we don’t waste time in case you are already way above the top range.”。“It’s a negotiation tactic but you aren’t actually negotiating salary at this point (or any point) with the recruiter.”。“A good recruiter aligns salary expectations on both sides before you ever reach interview. If the numbers don’t work, you shouldn’t be put forward.”。一位用人经理自己的台账：81 通第一通电话里有 51 通推进到下一步，最主要的两个淘汰原因是候选人失去兴趣，以及 “the candidate’s salary range expectations not matching what we can offer” ◆。

第二阶段，定 offer。 不同的机制、不同的人、不同的房间。按一家大型制药公司自己公布的招聘材料，“The recruiter will calculate an offer in partnership with the hiring manager and deliver the offer” ◆。在一家公开了流程的科技公司，offer 会进入一个有指定审批人的私有频道，超出准则的 offer “require approval from both the hiring manager’s manager and the Engineering executive” ◆。这层审批摩擦，正是“高数字在第一阶段让你付出代价、在第二阶段却几乎买不到什么”的结构性原因 ◇。

混淆这两个阶段的后果在证据里看得见。一位候选人报告了这个具体错误：“I brought up salary too early ; process slowed and eventually ended” ◆。而在同一份台账里，十二个流程中有两个死在薪资对话上 ◆。

这一点上证据确实有冲突，而且冲突是真的： 一位经验丰富的候选人坚持 “negotiations begin the moment you interact with a recruiter”，一家薪酬服务机构声称 “Negotiations can start in that first recruiter phone screen.”。与之相对，两位招聘人员直截了当地说筛选轮里没有任何谈判在发生。这个分歧对应的是说话人是谁 ◇：候选人和谈判教练把筛选轮当成锚定，招聘人员把它当成过滤。两者可以同时成立，因为一个能锚定的数字同时也是一个用来过滤的数字。实操上的结论是，在第一阶段你的数字只能让你出局或者让你通过；它不能给你挣到钱。

同时也要注意 offer 阶段实际能给出的空间，比筛选轮暗示的要大 ●：“I was negotiating an offer in the fall and was able to get the base 17% higher than their original offer, which was 20k over the range HR mentioned in the screening.”。“I said a number and got an offer over it.”。“my offers increased by 15% on average, the largest by 25%”。还有一个案例是第一通电话里说的下限最后变成了 offer：一位候选人说他需要 “minimum in the low-to-mid 300s in any combination of compensation components”，得到的回应是 “Okay, our ranges overlap depending on what you can bring to the table, let me schedule your final round”，最终以 306,000 成交 ◆。

这一条怎么用： 通话之前写下两个数字。一个是通关数字，也就是你会说出口的区间，选取标准是它可信地落在对方薪档之内，并且高于你真正的下限。另一个是下限，也就是你在 offer 阶段不会低于的数字，它在第一通电话里永远不说出口。在筛选轮里，你唯一的任务是在不给自己封顶的前提下通过那次比对。对自己说：在这个阶段我可能亏钱，但赚不到钱。

公布的区间往往不是真实预算

把这句本该只在私下说的话摆到台面上，因为一位招聘专员就是这么做的 ◆：

“the range is usually meaningless anyway. ive had comp teams hand me ranges like $120-180k for a senior engineer. that range exists because HR needed to file it with the state. the actual budget the hiring manager has? $145-155k. nobody tells the candidate that part.”

中文大意：公布的区间通常没有意义。薪酬团队给他的一个高级工程师区间是 12 万到 18 万，那个区间存在只是因为 HR 需要向州政府备案；用人经理手上的真实预算是 14.5 万到 15.5 万，而这一段没人会告诉候选人。

有候选人被直接告知过同样的事，构成独立佐证：“They told me that their budget was $90-95k and they were required to put that large range”，而公布的区间是 76,000 到 116,000 ●。还有候选人在通话中当场撞上这个落差：“They asked about compensation expectations and then provided a salary range that was lower and different from the range advertised in the job posting.”。“Especially annoying is when recruiters give you a range that could work, but realistically, the company will never give you top of range, and the real number is closer to the bottom.”。“MANY times I’ve asked questions about the JD and they tell me ‘that shouldn’t be there’, ‘that’s a typo’… even ‘no we’re not actually remote’ ‘no that’s not the true salary range’.”。

招聘人员描述了他们如何根据对你的判断，刻意报出一个收窄的区间 ◆：“If I have a role with a range of $100-130k, I gauge their experience and throw out a number that makes sense. If I feel there are not at the top end of I will say it is paying $100-115k and gauge their response.”。另一位把区间宽度本身点名为线索：一个红旗是 “The pay range is huge (more than $20,000 spread).”。

相反方向的做法也存在，值得知道 ●。有些招聘人员会公布并守住区间：“been in TA for 20 years and i just tell them the range. every time. first call. the ‘candidates will anchor to the top’ argument is nonsense… when you play coy with salary, you’re selecting for desperate candidates who’ll tolerate that.”。有一家公司把整个对齐环节交给一个公开的薪酬计算器，而不是从你嘴里套一个数字。有一家公司书面承诺 “Expect the recruiter to discuss the compensation bands with you during your first screening call.”。

这一条怎么用： 问那个现实的中位数，不要问区间。能问出可用答案的句子是：“I’ve seen the posted range. Where in that range are people actually landing for this role right now?” 公布的区间跨度很宽，是一个“该问这句话”的信号，不是一个关于你身价的信号。

Working scripts

下面这些句子来自那些报告了自己怎么用、以及结果如何的人。把每一句当成一个形状而不是一个剧本：里面的数字、职级和情境属于用过它们的人，所以在借用句子之前先把你自己的填进去。可迁移的是结构，不是数字。

只推一次，把问题交回去。 证据里被最多人认可、且报告有效的说法：“I don’t have a specific number in mind, what’s the range?” ◆ 一个更长的版本，还顺带表明你懂薪酬包的结构：“I’m open to the market range for this role. I understand there are a lot of different components within a salary such as base, equity, and bonus so I’m curious if there’s a specific range the team has in mind for this role.” ◆

回应他们刚给出的区间。 一位候选人报告自己用过几十次的三分支话术 ●： - 高于你的预期：“That sounds like the right ballpark, we can work with that.” - 与预期一致：“Hmm, is there room for negotiation in there? That would be the lower end of my expectations.” - 低于预期：“Thank you for sharing this upfront, those numbers are well below my expectations and so I’d like to end the conversation here, thank you for your time. Feel free to reach out if you revise your numbers in the future.”

同一位候选人报告第三个分支能当场逼出隐藏预算：“if they have the flex, this line brings it out right away. Just this week I used it and the recruiter on the other end said - ‘oh do you live close the greater NYC area? We can say that you’re in the New York zone and get you a higher TC’ If there is no flex, it saves you a ton of time” ◆。

另一位候选人给出一个中性替代方案，三种情况下用词完全相同，这样你的反应不泄露任何信息：“That range sounds like a good starting point,” 或者 “I think that’s worth having a conversation.” 而对于低得太离谱的区间：“It looks like this may not be a good match, and I don’t want to waste your time, so please feel free to reach out if the company decides to extend the range.” ◆

在必须给数字的时候给数字。 一位招聘人员自己的兜底话术：“Based on my research and the role’s requirements, I would expect a base salary in the range of $X to $Y. This range reflects my qualifications, experience, and typical market rate for this position.” ◆ 一位候选人的版本，带一个明说的退路：“Based on my research, the base for this position is $75k - $95k, and a number in that range would be acceptable as a starting point for negotiation, but contingent on the details of the position.” 他说明自己只在下面这种情况下才用它：“as a last resort if I’m getting the feeling that there is little chance I’ll be able to proceed with the opportunity if I don’t answer” ◆。

一位招聘人员给出的一组可用说法，适用于对方没有披露任何区间的时候 ◆：“I’m interviewing for roles that are in the $150-200k range base plus bonus and equity.” 或者：“I’m targeting a total rewards package over $250k with flexibility around the mix between base and bonus.” 或者，对于超出岗位要求的候选人：“I care a lot about internal equity so I’d like to understand your range better, but given that my experience exceeds what you’re looking for, I’d expect to be above the midpoint.”

只给下限，不给上限。 证据里唯一一条从第一通电话的措辞一路追踪到 offer 的成功记录 ◆：“I’m looking for a minimum in the low to mid 300s in any combination of compensation components to make a move.” 对方的回应是 “Okay, our ranges overlap depending on what you can bring to the table, let me schedule your final round”，offer 最终落在 306,000。注意这个结构：给的是下限，不是目标；“any combination of components” 拒绝了被钉死在基本工资上。

用公布的区间，但不主张它的上限。 “I’m comfortable with the salary range shown in the job ad, and if possible I’d like to discuss my salary expectation further later in the process.” ◆ 或者，在适用薪酬透明法的地方：“The listed salary range aligns with my expectations.” ◆ 一个更锋利、把活推回给对方的版本：“Could you narrow the posted band to the part that applies to someone at my level?” ◆

专门用来拒绝提供薪资历史。 这和期望薪资是两个不同的问题，值得一个单独的回答 ◆：“I’m not comfortable sharing proprietary information about how my current employer pays people.” 以及一个顶住对方声称的公司政策、有据可查的案例：“It’s my policy not to disclose my salary.”

在要价过高之后补救。 一位候选人向下重新锚定，流程因此继续而不是终止 ◆：他在对方说的 70 之下要了 85，被告知不可能给到那么高，接着他说自己在 75 也可以接受，面试随即被安排上。可挽回的动作是你自己立刻在同一次对话里报出一个更低的数字。

读懂岗位编制，并据此定价。 一位招聘人员点出了这个输入变量 ◆：“If the role has been open for quite some time the team will be more willing to move quickly for the right talent… If you have the niche experience they are looking for, raise your compensation target accordingly.”。在筛选轮里能问出这个信息的问题是：“How long has this role been open, and what has made it hard to fill?”

不要说的话。 “Market rate.”。“I’m sure you’ll make me a fair offer.”。“Whatever is the industry standard.”。这三句要么被招聘人员点名，要么有失败记录 ●。另外，不要在对方刚给出区间之后立刻主张它的上限；这是被引用最多的一个雷点 ●。

按赛道的决策表

这件事在漏斗的哪一层被决定，由谁决定

不是用人经理决定的，而且经常根本不是人决定的 ●。

最早的一层是申请表本身。一家公司书面写明：“Due to a high volume of Candidates, [we use] automated decision-making to filter Candidates based on responses to our application questions. When certain criteria is met regarding visa sponsorship, work permits, hiring country, and other employment restrictions, our software will reject your application.”。另一家公司的员工从内部描述了同一件事：“We have a question ‘are you currently authorized to work in the US’. ‘No’ answers are filtered out.”。担保这个问题和内推、既往雇佣经历等问题被放在同一个自动过滤的淘汰题块里 ◆。

第二层是招聘人员，而这里通常就是最终决定 ●。一家大型制药公司的用人经理：“We screen for visas via HR/ recruiter prior to me even seeing the resumes to think about which candidates to interview. So if youre getting to the hiring manager then its not the visa that is stopping you going further.”。另一家公司的员工：“Sponsorship is typically overseen by HR not hiring managers.”。一位招聘人员把它列为她不分岗位、对每一位候选人都必问的仅有三个问题之一：“if they’ll ever need visa sponsorship, what’s their availability timeline, and what salary are they targeting.”。

用人经理的热情压不过这一层 ●。“A few times, I called people who needed sponsorship and then asked HR to move forward with scheduling interviews. I always got pushback due to the visa issue. They really didn’t want to move forward even if the candidate had had a really positive phone screen with me. Eventually I learned my lesson and stopped doing phone screens for people who needed sponsorship unless we’d already failed to find someone who didn’t need it.”。在另一家公司：“candidates with visa requirement (including transfer) may pass HR and HM screen for the right qualifications, but would not proceed past that right now.”。

在大型科技公司还有第三道闸，法务 ◆。“All the recruiter is willing to tell me that the legal team has recommended against moving forward with the application and that he can’t go against their assessment.”。一位处在同一机制另一侧的候选人描述了这套流程运转起来的样子：“They all asked me to fill a small form so that their lawyers decide if my profile can pass it not.”。

这件事在漏斗中出现的时点并不保证靠前 ◆。有一位候选人先通过了筛选轮：“Everything went well and was expecting a positive response. But after a couple of days the recruiter called and told they are not sponsoring and even though I cleared the interview, she has to reject my profile.”。同一个帖子里有人说出了行业惯例：“If it’s during the first recruiter or HM call then yeah totally normal. Final loop, pretty weird and they should have figured this out by now.”。

注意招聘人员自己的交接表，它把这个答案以书面形式带到下一环节 ◆：表上把 “citizenship/sponsorship (date if needed)” 和薪资期望并列，作为一个常设字段。

这对你意味着什么： 那个可能被你的科学说服的人，通常不是决定这件事的人，而等你走到他面前时，决定早就做完了 ◇。不要围绕“靠打动人闯过这一关”来做计划。围绕“进入那些答案本来就是肯定的赛道”来做计划。

有据可查的一例：主动交代当场终结了流程

证据里有一个案例毫不含糊，也是最该记住的一个 ◆：

“My situation was a bit awkward. My OPT expires next February. I interviewed at a large search company and was told the interview results were fine, but when the recruiter asked whether I had any concerns, I stupidly said my OPT only had a few months left. The recruiter then went straight to the immigration team, who said they would not sponsor. They told me it was not moving forward, and my file was never even submitted to the hiring committee.”

中文大意：他的 OPT 次年二月到期。在一家大型搜索公司面试，被告知面试结果没问题，但招聘人员问他“有没有什么顾虑”时，他主动说了 OPT 只剩几个月。招聘人员直接找了移民团队，对方说不担保。流程就此终止，他的材料甚至没有提交到招聘委员会。

面试结果当时已经是正面的。被问的问题是 “do you have any concerns”，这不是一个关于移民的问题。候选人回答了一个没有被问出口的问题，而这个回答被送到了一个本来没有被咨询的团队手里。他自己的后续：“it just got stuck in a dead loop on the visa question”。

第二个案例以较轻的形式呈现同一种模式 ◆。一位候选人在回复一封安排时间的邮件时，未经询问就把完整的移民情况和盘托出：“I said”Thursday I’m flexible all day, I need visa sponsorship, transferring the visa only takes 15 days," and then she disappeared without confirming an interview time“。但这一例有第一例没有的结局：同一位候选人后来更新说”the recruiter got back to me. The reason she had not replied was that she was discussing it with the hiring manager, and they decided to skip the phone screen and send me straight to the hiring manager interview"。披露之后的沉默不等于被拒 ◆。

这一条怎么用： 回答被问到的那个问题，然后停住。“Do you have any concerns?” 不是签证问题。“when could you start?” 也不是，除非你有一个日期会让这个回答变得不准确，那种情况下你给出日期，不给理由。

另一侧有据可查的代价

回答不实的证据同样是第一手的，而且失败发生得很晚，在你已经把一切都投入之后。

最具体的一份记述 ◆：

“Seven or eight years ago I concealed it. Two companies had already shipped the package to my home, including tax forms and company credit card applications. Then when I came clean about when the H-1B transfer would happen, both of them immediately sent me a UPS overnight label the next day and asked me to return the package.”

中文大意：七八年前他隐瞒了这件事。两家公司已经把入职材料寄到他家，包括税表和公司信用卡申请。等他坦白 H-1B 转换的时间点之后，两家第二天就寄来 UPS 隔夜快递单，要他把材料退回。

另一位候选人，不同年份、同一个问题：“The other company that also gave me an offer destroyed the offer because of my”yes" answer."。这一条是如实回答的代价，不是不实回答的代价，它应当被记录在案 ◆。

招聘方把同样的结果描述为政策 ●。“If someone responds ‘no’ to the question ‘will you require visa sponsorship now or in the future’ and it comes out during the offer process that they lied, that can be reputation ruining.”。“If you lie and make it to the offer stage, they’ll find out then and pull your offer.”。“Do you have them answer as part of the application? Record of their application and then if it gets further in the process and they reveal they need sponsorship, they’ve now lied on their application and that’s an integrity issue.”。还有一位实际执行筛选轮的人说到这种失败方式：“Claiming no immigration sponsorship required when applying, then ‘clarifying’ your ‘mistake’ on the screener… I promise that does not work nor will it ever, so stop doing it.”。

招聘人员现在也预期到这一点，并直接针对它筛查 ●。“Just had a call with a recruiter and he said first thing he needed to verify is that I’m a US citizen and don’t need sponsorship and that he’s met with lots of people the past week who said they were a citizen in their applications but once they meet they say they aren’t citizens and need sponsorship.”。从招聘人员那一侧：“lately candidates have been getting passed my screening question asking if they require sponsorship or will in the future. Most lie and say no, when they actually do.”。以及：“If I have a feeling they’ll need sponsorship it’s one of the first questions I ask on the screen and quickly end the call if they say yes.”。

即使是如实回答，说晚了也另有一层较轻的代价 ◆：“When candidates don’t tell us about sponsorship requirements, which is part of our application process, it’s an orange flag. We always try and sponsor where possible but withholding that information is never a good start to employment.”。

这一条怎么用： 你有一条规则可以了结这件事，而证据在实效层面而不只是在原则层面支持它。不实的回答买不到工作；它买到的是一个更长的流程，终点是 offer 阶段被撤回 offer ●。每一次被问到都如实回答。你的空间在对方问了什么里，不在事实是什么里。

雇主是怎么把模糊性堵死的

候选人试图利用的那道缝，是“现在有工作许可”和“以后会需要担保”之间的差别。雇主已经用三种各自独立的方式，以书面形式把它堵死了 ●。

第一，他们问两个问题而不是一个，而且第二个是将来时。候选人报告到处都会遇到的那一对问题，逐字如下：“1) ‘are you legally allowed to work in the United States’ 2) ‘Will you now or in the future ever need visa sponsorship’.”。一位招聘人员把后果分开说明：“If you answer No to question 1 you will be rejected. We cannot work with you. Legally… Question number 2. if you answer Yes, it is not an immediate rejection. This means you are on a work VISA and require extra steps before we can hire you. It does lower your chances as some companies will not sponsor you so your options are more limited, but it’s not an immediate no.”。这个区分，硬过滤对软过滤，是本节最有用的一句话 ◆。

第二，一家大型科技雇主在表单上明确列出了具体身份，让文字游戏没法玩 ◆：“To stop candidates playing word games, after asking this question IBM adds a line saying that if you are on OPT, CPT, STEM OPT, or H-1B, you must answer”yes“.”。另一家把问题写成把整个未来都包含在内 ◆：“If hired, will you need immigration assistance or sponsorship from Facebook to begin or, at any time in the future, to continue to work lawfully for Facebook in the United States?”。

第三，也是对一位持 OPT 的博士最重要的一点，筛选轮会重新问一遍并重新解释。一位生物技术公司的用人经理说出了确切的顺序 ◆：

“As a heads up, if you answer no to that question, our next step is an HR screen where we will ask if you’re on OPT. To us that counts as sponsorship and if we’re unwilling to sponsor, we won’t hire anyone on OPT either.”

中文大意：提前说明一下，如果你对那个问题回答“否”，下一步的 HR 筛选轮我们会问你是不是在 OPT 上。对我们来说 OPT 也算担保，如果我们不愿意担保，那我们同样不会录用任何在 OPT 上的人。

这里存在一个真实的分歧。一位候选人转述了社区里占主流的看法：“as long as you have current legal working authorization you are fine to put down that you don’t need it and discuss the issue at the interview” ◆。上面引用的那位用人经理正是对这个看法的直接反驳 ◆。分歧对应说话人是谁：候选人描述的是什么能让他们拿到面试，雇主描述的是筛选轮上会发生什么。对你个人而言，这个分歧没有意义。 问题里有将来时的从句，表单上又点名了具体身份，对一个持 OPT、将来会需要担保的人来说，不存在一个如实的“否”。所以这个问题根本不成立。

而且不存在一种能绕开这个问题的申请方式 ●。一位人力资源总监：“you need to be consistent in asking it ie ask the same question to everyone. Ideally on your application form and again in your pre screen. Yes, you can ask candidates if they require sponsorship to work in the U.S., as long as the question is phrased appropriately and consistently asked to all candidates to avoid discrimination.”。“This is usually a standard question on every application.”。

这一条怎么用： 不要再把这个答案当成一个变量。它是固定的。你的全部余地在于你申请哪些雇主，以及你如何界定担保对他们意味着什么，也就是接下来两节的内容。

cap-exempt 身份：它能做什么、不能做什么，以及时间账问题

它是什么。 一所大学的国际就业办公室列出了适用类别 ◆：“Employers that fall under the following categories are able to sponsor an H-1B visa without being subject to the annual cap: Higher education institutions; Nonprofit entities that are related to or affiliated with a higher education institution (such as hospitals); Some Nonprofit research organizations and Cap-subject organizations that require the H-1B employee to work exclusively with one of the above categories of cap-exempt employers.”。同一来源指出它并不限于研究人员：“You can receive an H-1B for many different types of specialty-occupation positions that support qualifying organizations, as long as the organization is willing to sponsor your visa.”。

它能做什么。 它去掉了抽签。一位候选人在给一名在国家实验室和半导体公司之间做选择的国际博士生建议时说 ◆：“National labs don’t have h1b quotas so if you get an offer there, as long as you are eligible for an h1b, you get ypur visa in 2 weeks. Even if [the company] wants to hire you, they have to apply for the visa lottery if you will need h1b at some point.”。

它不能做什么，而这才是最要紧的部分。 cap-exempt 身份属于雇主，不属于你 ●。它是“谁在提交申请”的属性，不会随你移动。“Academic H1Bs are cap-exempt as well, so if you want to transfer to a capped H1B at some point, you will still need to go through the H1B lottery.”。“you can’t really move to cap subject later on and it’s really hard to jump to another cap exempt. No company will file lottery for you even if you get a job in January in case you are wondering. Ultimately i left US and moved to EU.”。“They cannot move from a cap exempt h1B to cap H1B so easily.”。

它也不代表雇主真的会用这个身份 ●。“Many of these institutions have a lot of roles that ‘don’t qualify for (visa) sponsorship’. This means while they have jobs open, HR may block them from sponsoring H-1B (even cap exempt) by organization policy.”。一位在 cap-exempt 医院系统的候选人报告说，一个 principal 级别的岗位让他两次确认身份：“they made me confirm my legal status twice over because they weren’t doing anything for anyone anymore. Wouldn’t even do the non-lottery, non-$100K cap exempt COS H-1B for a PhD principal position. Immigration matters are just radioactive right now” ◆。而在当下的经费环境里：“getting cap exempt is very difficult or almost impossible after research grant cuts by the Government. You can try but don’t have much hope” ◆。

时间账问题。 这是整道闸里最重要的一段分析，它来自一位雇主和一位候选人之间的一次交锋 ◆。

这位雇主来自一家 cap-exempt 的非营利研究机构，说明了他们的政策：“We never commit to sponsoring a green card in the hiring stage. When an employee is in year 2 of their SECOND H-1B, we will then look at sponsorship, many have already applied for EB2 NIW or EB1 OR. It also depends on funding. If a lab is going to lose funding, we take that into consideration too.”。

回帖把这笔账算了出来：“they said they start looking at sponsoring GCs at year 2 of the second H-1B (i.e. ~5 years into the H-1B). The problem of course, is that even getting to filing PERM takes 1.5 years nowadays, and you can only extend H-1B if the PERM was filed at least 1 year before H-1B timeout. This is in fact a shining example of how large non-profits/research organizations don’t understand immigration timelines, are reticent to sponsor, and by the time the bureacratic organization is willing to sponsor, it is often too late.”。

反对意见的其余部分 ◆：cap-exempt 雇主 “tend to pay a lot less than industry, up to 50-75% less”，而且 “often don’t sponsor green cards for many roles due to expense and bureaucracy. Once 6 years elapse, you may have to leave the US.”。

它仍然是正确选择的条件，由同一位评论者给出 ◆：“(1) H-1B sponsorship is readily available, (2) there is a clear and viable path to a green card post H-1B, (3) your country of birth is not stratospherically backlogged (sorry, India and China).”。

以及那个不同意见，它是第一手的，不该被丢掉 ◆：“I work for a cap exempt employer; always have. Yes most cap exempt don’t pay as great as cap subject employers but that you get job stability, less visa issues and great work life balance… personally I feel that it’s kind of an advantage to someone like me with less competition!”。

这个分歧对应的是绿卡路径，不是 H-1B ◇。所有人都同意 cap-exempt 让 H-1B 变得容易。称它为陷阱的，是那些永久居留路径卡住了的人；喜欢它的，是那个没有在跟时钟赛跑的人。

这一条怎么用： 如果有一个 cap-exempt 雇主在考虑范围内，接受之前该问的问题不是 “will you sponsor an H-1B.”。而是：“What is the institution’s practice on permanent residence sponsorship, at what point in employment does it start, and has the office filed PERM for someone in a role like this in the last two years?” 一位候选人把这条规则说得很直白：“if you are considering a cap exempt job you need to have a crystal clear understanding of their stances regarding immigration support… And because of the limiting nature of a cap-exempt H-1B, you need to know this ahead of time. With a regular H-1B you can always find another employer if your employer drags their feet on PERM.”。

关于什么时候提出来的分歧

这是一个双方都有第一手证据的真实冲突，靠选一个赢家是解决不了的。

在第一通电话里提出来。 某个帖子里 2026 年的共识 ◆：“Should bring up in the phone screen before any interviews. You maybe wasting your time or their time if it already has been decided. If you think they will be blown away and say that they must have you and therefore will sponsor, dream on. That doesn’t happen in this current environment… If they don’t ask in the application, bring it up during the first recruiter/HR call. Be transparent early, but don’t make it the first thing they see.”。其他帖子里的候选人说法相同 ●：“If I need visa support I obviously have to make it clear from the start. Otherwise I interview through a whole round and at the end they do not sponsor, which wastes both sides’ time.”。以及：“I try my luck and I usually bring it up in the first round of interviews to save us both some time.”。招聘人员从他们那一侧表示同意：隐瞒这件事 “is never a good start to employment.”。

等等再说。 一位候选人自己的统计是另一种立场最强的证据 ◆：

“i) I currently work at a F500, and the only reason I got in was because I didn’t specifically say I needed sponsorship during the recruiter screen. I went through the interview loop; the HM liked me a lot and he was okay with my visa status when I told him later. ii) I did a lot of recruiter screens last year and got rejected from most of them after I said I needed sponsorship. The week before Christmas I got instantly rejected from two recruiter screens in a row after I told them I needed sponsorship.”

中文大意：一、他现在在一家世界五百强工作，能进去的唯一原因是他在招聘人员筛选轮里没有特意说自己需要担保。他走完了整轮面试，用人经理很喜欢他，他后来告诉对方签证情况时对方也接受了。二、他去年做了大量招聘人员筛选轮，说了需要担保之后大多数都被拒；圣诞节前一周，他连着两次在告知需要担保后被当场拒掉。

要仔细看清他做了什么、没做什么。他没有说自己不需要担保。他也没有主动交代自己需要。他回答了被问到的内容，然后在用人经理那一环节披露。那就是那条窄路，而它恰好就是你那两条规则所描述的路。

这个分歧对应什么 ◇：它对应雇主问不问。凡是在筛选轮里就会被问到的情况，而按证据这几乎总是会被问到，这个分歧就是空的；你如实回答，时点就已经替你定好了。凡是雇主不问的情况，分歧是真实的，而两种选择的代价在性质上不同：早说，代价是失去那些本来就不会转化的流程；等着说，代价是在那些本来就不会转化的流程里损失好感和时间。两种做法都无法把一个“不担保”的雇主变成担保。

还有一种立场值得记录 ◆：“You can include your visa status in the resume to avoid disappointments in the future.”。这是披露程度光谱上最大的一端。对一个优先节省时间、而不是优先争取见到用人经理的人来说，这是一个正当的选择。

这一条怎么用： 不主动交代，也不回避。如果申请表问了，就在那里如实回答，并把它视为整个流程中你的披露。如果没有任何环节问起，而你走到了一位热情的用人经理面前，那就在现场面试或专家小组面试被安排之前自己提出来，因为一场排定的小组面试是沉没成本开始转到对方身上的地方，晚披露的好感代价从那时起变成真实的 ◇。

四个赛道上的差别

小型研究机构。 在当下的证据里是最不友好的赛道，原因和你本人无关 ●。一位用人经理：“We are told by HR to strongly prefer local and natives (anyone who would not need sponsorship).”。另一位：“visa hiring requires you show you did not have similarly skilled US applicants which is very hard to prove with current market conditions.”。经济账被具体点名 ◆：“The STEM OPT is not the issue, it’s what comes after. Unfortunately it’s an expensive process to sponsor somebody and not be guaranteed they get selected in the H-1B lottery. Last year an expedited H-1B transfer with lawyer fees cost around 9,000 at my org.”。

但这个赛道也有唯一一种科研团队能强行改变结果的机制 ◆：“The hurdle has largely been that companies are on a hiring freeze and HR isn’t interested. But if the team says they need someone to do a specific role or need to hire they do still have the ability to bring about an o1.”。走通小型研究机构这条路，靠的是科学家们点名要你，不是靠招聘人员。

大药企。 政策形式上相同，但有一个明说的门槛 ●。“None of the big pharma/giant biotech companies even look at your resume if there’s a visa sponsorship criteria checked yes.”。招聘方给出的理由：“Why would I go through the expensive and time consuming trouble to sponsor a visa holder when I can hire a citizen or permanent resident with the same qualifications? You will only see companies sponsoring for highly specialized skills and high levels (execs).”。一位用人经理点出了实际的标准线：“Wouldn’t do it for candidates below 95% percentile because immigration’s bureaucracy at the moment such a pain.”。另一位描述了招聘信息里不作任何声明的静默过滤：“My department is hiring right now and we outright reject anyone needed visa sponsorship… We still sponsor for higher level position but this is entry level PhD.”。

反例是真实存在的，而且形状完全一致 ●：一项小众的、被点名的、难以替代的技能。“I met a European postdoc who got H1b sponsorship from [a large pharmaceutical company’s] protein science division a few months ago, so opportunities are still out there if you have a very niche skillset that they desperately need.”。“I know companies that are still sponsoring O-1s (think AIBio PhDs with citation counts in the thousands). H1-Bs not so much.”。有两位员工各自独立地点名了同一家公司仍在积极担保。

研究机构、国家实验室、博后。 结构上最可得的赛道，也是最可能把你锁住的赛道 ●。机构的默认做法比你的谈判更重要。大学出于一个明说的经济理由推 J-1 ◆：“One reason universities won’t do H1Bs out of the gate is they have limits on postdoc length. Usually something like 4-6 years and the visa can be extended to 5. So why would they pay a ton more cash for an H1B when you can fulfill your entire contract on a J1?”。有些机构有明文政策 ◆：“Some institutions will have policies (‘soft’ or ‘hard’) about favouring J1 sponsorship so won’t sponsor a H1B if a J1 is an alternative. J1 is a lot cheaper for the sponsor so they do tend to push it, especially in the first 1-2 years. Some PIs also know that postdocs that get a H1B, then EAD/Green Card won’t be as likely to stay around, so don’t want to speed up the process.”。

同一个帖子里的反方论点 ◆：“Academic H1Bs are NOT capped and never participate in lottery (as long as the position requires a master’s or higher, so all postdocs and faculty positions by definition), so, if an applicant manages to negotiate H1B, they’re better off going for it over J1 if they want to stay in the U.S. afterwards. J1 might require returning to your home country, and getting a waiver is never guaranteed.”。

据报告国家实验室在做同样的事 ◆：“Most national labs have become clever and simply utilize the F1 OPTs or J1 visas to hire ‘Postdoc students.’ Once you spend some time here, you essentially lose chances to partake in H1B visa lotteries.”。同一位评论者点名了哪些实验室在录用非公民方面更容易，并建议直接向招人的 PI 提出这件事，“straight but subtle and politely.”。

一位实际完成转换的人给出的成本和时间线 ◆：“I got a PI to sponsor me for a H1B after my J1 hit the limit and changed labs. The univeristy pushed back a little, but he argued the case. It costs about $5k, lawyers, and time. Takes about 2 months with fast tracking. So it is effort they would rather avoid if possible… Most PIs are clueless about international visas so you have to be explicit early on as not to supprise them.”。

以及一条关于陌生邮件的警告 ◆：“if there are sponsorship requirements AND I am not familiar with your advisor or yourself, then there is no chance cold emailing works.”。

大厂。 最系统化的一个赛道，在证据里也是对特定路径最开放的一个 ●。过滤是自动化的，而且精确，它区分转换和新担保 ◆：“Google and all other big tech have a question on their online form… If you don’t have a visa and you don’t currently live in the US and you require sponsorship to get an H1B, that triggers a rejection… Big tech famously hires H1B transfers and other visas but not new H1Bs.”。

多份记述都把 O-1 当作一条独立且更受欢迎的路径 ◆：“last year when I was interviewing at big tech, they all agreed to sponsor an O-1. They all asked me to fill a small form so that their lawyers decide if my profile can pass it not… I haven’t yet met a recruiter who said no to O-1, they prefer it to H1B. The only thing is that you need to clear the bar.”。与之相对，一位在同一类公司的候选人报告了在 O-1 上相反的结果：“the recruiter asked if I needed visa sponsorship and said that they couldn’t provide it for an O-1 visa.”。两者都是第一手的；这个分歧似乎对应的是律师评估下每位候选人 O-1 案子本身的强度 ◇。

在美国以外，同样这些公司把它当成例行公事 ◆：在一家大型研究实验室，“if your concern is about getting a visa, that’s relatively easy to do. [We] will provide assistance with the process. I went through it myself and it was relatively painless; it probably took 5-10 hours of my time total.”。另一家公司的招聘人员给它标了价格和回款条款：“it costs the company $10k for a visa sponsorship to the UK and if this role left before the 2 year mark, the employee has to pay some of the fees back” ◆。

Working language

下面这些句子是形状，不是剧本。它们是为一位现在有工作许可、将来会需要担保的计算科学博士写的，其中没有一句提供了对方没有问的信息，但每句话里点到的身份都必须和你自己的事实一致，你才能说出口。说之前先把每一处方括号的内容对照你自己的情况核一遍，因为这道闸的全部价值就在于你的回答是准确的。

申请表问那两个标准问题时。 按问题本身回答。“Are you legally authorized to work in the United States?” 如果你当前持有效 OPT，答案是 Yes。“Will you now or in the future require sponsorship for employment visa status?” 答案是 Yes。不存在另一个准确的选项，而且证据显示，即便“否”能让你拿到筛选轮，它也撑不过筛选轮 ●。

招聘人员在电话里问起时。 完整的回答，说完就停：

“Yes. I’m authorized to work in the United States right now, and I will need sponsorship at some point in the future.”

如果你想加上那一个能改变雇主成本核算的事实，而且只在你想加的情况下，就加下面这一句，然后停住：

“Yes. I’m authorized now, and I will need sponsorship later. Happy to walk through the specifics with whoever handles that on your side.”

第二句话把细节交给了对的职能部门，而不是交给一个用不上这些细节的招聘人员 ◇，而且它不会引着你去讲述自己的 OPT 倒计时。

当他们问出一个你可以正面回答的追问时。 回答这个追问，不要回答它的放大版。

• 如果被问是哪种身份：“I’m on [the status you actually hold].”
• 如果被问你是否需要走 H-1B 抽签：用事实回答，并且只在适用时才提到豁免。“For a cap-exempt employer, no lottery is involved. For a cap-subject employer, yes.”
• 如果被问到你确实符合条件的其他路径：“I would also be a candidate for an O-1, and I’m happy to have your counsel assess the case.” 这是证据显示能在大型科技公司打开一扇门的句子 ◆，而且只有在它对你的履历为真时才可用。
• 如果被问什么时候能入职：给出日期。不要讲为什么。

当他们问的不是一个签证问题时。 “Do you have any concerns?”、“Is there anything else we should know?”、“What’s your timeline?”。这些都不是移民问题，而证据里就有一个人把它们当成移民问题来回答，结果在材料送到委员会之前就丢掉了一个本来正面的流程 ◆。就事论事地回答它们：“No concerns. I’m interested and available.”

当你被直接问到公司是否需要新办一份 H-1B 时。 如果你在 OPT 上，代价最小的如实表述方式是证据里点名为更容易被接受的那一种 ◆：雇主觉得转换比新担保更容易。如果你的情况确实是转换，就直说。如果不是，就不要暗示它是。

在答案还没有定下来时，向他们提一次问。 用一种能问出政策答案而不是个人看法的形式，把问题交给招聘人员：

“What is the company’s current practice on sponsorship for this level of role?”

而在 cap-exempt 雇主那里，在任何 offer 之前：

“Does the institution sponsor permanent residence, at what point in employment, and has that happened for someone in a role like this recently?”

绝对不要做的事。 不要对将来是否需要担保的问题回答“否”。这对你来说是不实的，它会被查出来，而证据显示查出来的时点就在 offer 阶段，在一切都已经投入之后，在入职材料都已经寄出之后 ●。不要在一个并没有问到这些内容的问题下，主动交代你的 OPT 到期日、你的抽签历史、你家人的身份，或者你的长期计划 ◆。还有，不要把如实披露之后的沉默当成拒绝；至少有一个有据可查的案例里，那段沉默是招聘人员在安排跳过筛选轮 ◆。

1. 这一轮为什么不一样

第一轮非技术面通常是招聘人员，而招聘人员在结构上就无法回答你想知道的大部分内容。一位招聘人员直接说明自己的边界：“I usually share that I’m not on the team so I can answer general organizational questions or logistical questions.” ● 一位亲自面过数百名候选人的用人经理从另一侧画出同一条线：“ask your recruiter those questions, not the interviewer. I have interviewed 100s of candidates personally and you can only comment about your own working hours and the flexibility of your team, not the company… The questions about wfh policy and org structure are best left to the recruiter.” ● 还有一位顾问写给一名申请资深科学家岗位的生物化学博士：“Typically phone interviews are with HR people, so they won’t know much about technical details.” ◆

所以，一位称职的招聘人员能回答什么，是有一条底线的，而且有招聘人员把这条底线讲了出来：“Any decent recruiter can speak confidently at a high level about what you’ll do in the role, the team size, the cross-functional partners, and some sort of a growth trajectory. If they can’t give you those basic things, run. The company does not know what it wants.” ◆ 最后那半句才是可执行的指令。当招聘人员连自己职责范围内的问题都答不上来时，这个答不上来本身就是发现。一位候选人问，什么样的人适合这个团队，得到的回答是 “someone who does it the [company] way”，这对一个本该由招聘人员负责的问题来说，是零信息量的答复。◆

时间也很短。一位招聘人员描述的筛选轮结构大致是：十分钟公司概况，十分钟契合度提问，两三分钟事务性内容，剩下的才归你。另一位说整通电话是 “10-15 min in my experience. Depends if you have questions for them.” ● 第三位干脆把候选人提问放到最前面：“I let candidates ask their questions AT THE START OF THE INTERVIEW since that always ends up leading to the pitch anyway.” ◆ 按三到五个问题准备，不是十二个；并且要假定提问机会可能出现在你还没听到任何信息之前。

有两点让这段时间的价值超过它的分钟数。第一，招聘人员答不上来的问题仍然会被带走：“sometimes they answer and you can tell how connected they are to the job, other times they say ‘that will be a better question for the hiring manager’ If they like my question in either case, they may take special note and pass it along to the next person. I had an interview the other day where the hiring manager was following up on the recruiters notes from our conversation.” ◆ 第二，招聘人员手里握着别人不会给你的东西，也就是用人经理的需求交底。一位候选人这样描述真正的收获：“a ‘sneak peek’ of the key points the hiring manager noted, basic information on the team/structure, and typically important items that will be in the first few months.” ◆

一个结构性警告。一位招聘人员描述她如何结束一通已经决定否掉的电话：“As soon as I realize that they’re not a fit, I stop asking questions and say, ‘that’s the last question I have. We’ve still got a minute. Do you have any questions for me?’” ◆ 非常早、非常突兀地把话头交给你提问，是关于你处境的信号，不是邀请。◇

2. 有效的问法

按它们能换到什么来分组。每一条给出的都是实际被说出或写下的原句。

3. 产生反效果的问法

4. 阶段门

5. 按对象区分

同一个问题落在不同的人手里，效果不同。收尾处那族关于顾虑的问题是语料里最清楚的案例，值得把证据到底显示了什么说得精确一些。

在就这一族问题发表意见的招聘人员当中，负面多于正面，而且举出具体损害的都在负面一侧：“comes across a little desperate. Don’t ask it”；“I hate all variations”；“they don’t show your understanding of the role or the business”；再加上那份包含 “How’d I do?” 的烦人清单。有两位招聘人员确实喜欢它，其中一位还主动给出了更柔和的改写。在用人经理当中，模式在性质上不同：两位描述了自己被这样问过的具体实例，两位都报告它奏效了。“One great question someone asked me was ‘do you have any concerns that I can address?’ I thought this was brilliant because I did circle back to some of her gaps. It gave her an opportunity to further address with me.” ◆ 以及 “I always ask what they think a successful candidate will be expected to accomplish in the first 6-9m, what the profile of their ideal candidate looks like, and if they have any concerns or things I can clarify from my resume/the conversation.” ◆ 语料里用人经理提出的反对，表述方式是对套路问题的政策性偏好，不是候选人受到损害的记述。

由此得出的实操规则是：面对招聘人员，用那个收窄的重答式措辞，或者干脆跳过这一族；面对用人经理，关于顾虑的问题是可以放上桌的。◆

有一个来源给出了按对象分配的明确方案，值得整段引用，因为它是这里独一份的：“For potential colleagues in the same group, ask about the culture and what they like/dislike. For potential colleagues in a different group, ask about how the communication is between the different groups. For the hiring manager, ask what the expectations will be and what stresses them out the most. For upper management, ask them where they see the company a year from now. For HR, ask general questions about culture, company events, etc.” ◆

关于创始人这一列有两点说明。一位创始人确认这些问题是被打分的：“people who are very interested in joining a company have done their homework. They typically have several detailed questions, and those questions reveal a lot of things those people find important in their job.” ◆ 另有一位创始人告诉你他反过来在拿什么给你打分，你应该准备好回答：“What is your appetite for risk? There is a high chance of failure.” ◆ 一个生物技术领域的来源在创始人这一轮的钱问题上持相反立场：“Dont ask much about funding or investors or negative structural changes. They dont like that much from my previous talks.” ◆ 这个分歧是活的；见第 7 节。

6. 唯一的铁律

一定要问点什么。一个问题都不问，是整个语料里被点名次数最多的红旗。 ●

它作为明确的拒绝理由出现在一封逐字的拒信里：“You had very little questions about the company and the interviewers didn’t feel a lot of interest from your side.” ◆ 它在一位用人经理那里是头号条目：“The biggest red flag for me is when a candidate asks zero questions.” ◆ 一位生物技术面试官：“It’s very much a red flag to not have any questions.” ◆ 另一位：“If you have nothing to ask, I will assume you are uninterested and didn’t research the company or role.” ◆ 一位招聘人员：“The questions you ask tell me how you think. They tell me whether you’re already picturing yourself in this role or just hoping to survive the conversation.” ◆ 一位招聘人员讲基线效应：“The fact that you are asking any questions is a great sign… someone who acts like a dead fish” 是另一种选项。◆ 一位人力资源从业者在列举候选人在电话筛选轮绝不能说的话时，把它和另外三条并列：“No questions to ask the company.” ◆ 关于它怎么进入复盘会：“If you simply say you have nothing to ask, unless there are twenty or thirty minutes left and you have genuinely exhausted your list, it can become a yellow flag in the debrief.” ◆

我承诺过不留反例，而语料里有两条，所以列在这里。一位生物技术面试官持异议：“a person running out of questions to ask was never a big deal, and if anything just felt honest.” ◆ 一位面试官说这个环节根本不计分：“When I interview someone, asking no questions or asking a silly one does not cost points, but asking a good one earns them.” ◆ 两条都是少数派立场，对面大约有八个来源；并且请注意，两位在否认下行风险的同时，都承认上行收益是真实的。还有一条双方都接受的调和说法在案：“Running out of questions is not the same as not having any questions.” ◆ 在流程中途把清单问完是可以的。空着手来不行。

最低限度的两个问题，用于通话拖长、你只剩三十秒的时候。 一位招聘人员把它表述为一条明确的下限，并且两句都给了出来：

“Have at least two questions ready. Why is this position open is a good one. And it tells you if someone got promoted, quit, or was fired. And that’s useful information for you. What does success look like in the first 90 days? This question shows that you’re already thinking about how to deliver.” ●

中文大意：至少准备两个问题。“这个岗位为什么空出来”是个好问题，它会告诉你上一个人是升职、辞职还是被辞退，这对你有用。“头九十天的成功是什么样”这一句显示你已经在想怎么交付了。

这两句，按这个顺序，就是全部的兜底方案。它们是招聘人员答得上来的，它们都出现在多份互相独立的推荐清单上，并且合在一起覆盖了你在进入下一轮之前最需要的两件事。

7. 未决

这些分歧是真实存在的。我把它们记录下来，而不是选边。

提问环节是否被计分。 一位招聘人员说基本不：“recruiters often aren’t evaluating you too much based on the content of your questions, but they do notice the impression you give.” ◆ 另一位说决定早就做完了：“By the time we get to ‘Do you have any questions for me?’ I’ve already got a good sense of how I feel about the candidate.” ◆ 与之相对，一位从业十年的招聘人员：“At the end of an interview, when I ask ‘What questions do you have for me?’ I have already decided how I feel about your skills. At this point, I want to understand how you think.” ◆ 还有一位用人经理，他给出的最高票答案是，能让他印象深刻的正是那些搜寻红旗的问题。◆ 安全的读法是：下行风险很小，上行收益是真实的，而这恰恰是双方实际描述的内容。

问题的预算。 一位主持开放提问场次的招聘人员持最激进的立场：“The best questions you should be asking in an interview don’t really change depending on the stage… you should be asking strong questions the entire way through. Don’t hold anything back thinking you’ll save it for the next round, because the next round isn’t guaranteed.” ◆ 有两位候选人被拒，明确反馈是“问得太多”。◆ 第三个来源建议固定三个问题的结构：一个技术问题，一个关于团队和管理的问题，外加一个轻松的备用问题，时间不够就砍掉。◆ 我无法裁决这一点。这个可砍掉第三问的三段结构，是语料里唯一一个在两侧立场下都站得住的方案。◇

“based on this interview, do you think I’m a good fit for this role?” 有一个第一手案例在它之后拿到了 offer，至少一位招聘人员称它显得急切，一位称它可以接受但很弱，一位喜欢它。●（指分歧确实存在）未决。

筛选轮问福利。 多位工程师和一起第一手成功案例说第一通电话就问。一位从业三十年的招聘人员说等到他们想要你的时候。● 未决。注意这两个立场说的是不同的事：你有权看到什么，与你的筹码在什么时候达到峰值。

创始人那一轮问 runway 和资金。 同一场讨论里的两条评论站在对立面，一条说 “Dont ask much about funding or investors”，另一条说 “Ask how much runway they have. Do they have series A or B funding?” ● 一位 Series A 之前的创始人认为这整类问题对候选人几乎没用：“Asking about growth plans, runways, product market fit, Founder experience, etc. is so hard to judge as to be hardly worth it… Just: do I want to work here and how big an issue is it if I lose my job in the next 3 to 18 months.” ◆ 另一位创始人则乐于回答。◆ 贯穿所有立场的唯一稳定发现是：反应比数字更有信息量。

签证担保的时机。 一份第一手统计显示，在招聘人员筛选轮披露带来了当场被拒，而拖到用人经理阶段才说则换来了一家大公司的工作。◆ 一位招聘人员称隐瞒是橙旗。◆ 时间最近的共识说第一通电话就提。◆ 由于招聘人员反正会普遍问到这一项，这些立场之间的空间比看上去更窄，但当你确实有得选时，哪一种更好，我不做裁决。

算力分配、发表与开源政策、实验室权限。 没有。这份证据基础在这三件事上既没有有记录的问法，也没有有记录的结果。如果你想要针对它们的措辞，那不会来自这里。

1. 这一轮到底是什么

创始人轮，是和那个为结果负责的人谈：创始人、首席执行官、实验室主任、PI。在小机构里，它常常不是最后一轮，而是第一轮，有时候还是唯一的一轮。一位创业公司创始人描述自己亲自跑初筛：“as a first step do a quick 10-15 minute phone screen. If the candidate is a poor fit - you only spent 10 minutes on them.” ◆ 一位博后申请者报告的是同样一种“中间层不存在”的情况：“I finally have gotten the opportunity to interview with a PI with no ‘pre HR interview’.” ● 在一家国家实验室，“the host PI and core project members joined the interview”，之前完全没有 HR 筛选轮。●

关于这一轮，有两种说法流传很广，而它们都在产生这些说法的同一批讨论里被推翻了。

第一种说法是：这只是一次文化闲聊。“Mostly CEO interviews are a vibe check，”一位评论者说，另一位附议：“CEOs being the last round is fine and is usually just a culture fit / casual round.” ● 这有时候确实准确。一位候选人报告，“the CEO interview round consisted of talking about which restaurants we’d been to in London”，而且照样拿到了 offer。◆

第二种说法是：这只是走个形式。某个帖子里票数最高的评论就是这么定性的：“They’re not sending everyone to the CEO so this is probably one of those deals where the CEO likes to talk to EVERYBODY before hiring and he’s not going to overrule the hiring manager unless you totally flub it.” 另一位说得更硬：“These CEO rounds are nigh on pointless but small companies are in love with them. They’re usually softballs and if you don’t make a fool of yourself they’re not going to block you.” 这两条都被标为有争议，理由就在同一批帖子里。一位候选人：“Talked to literally 6 people and the CTO and CEO after a meeting where quite frankly nothing happened decided to override everyone and not hire me.” ● 另一位，在三轮“100% positive”的反馈之后：“He vetoed my offer due to my experience not being deep enough.” ● 第三位在系统设计讨论进行到十五分钟时被砍掉：“then he just abruptly ended it with a ‘have a nice day’ after 15 minutes. No feedback.” ● 第四位：“in founder round, I got rejected in between my intro within 3-4 mins as I don’t have B. Tech.” ◆

真正准确的描述来自一位在这类流程内部工作的评论者：“Every place I’ve worked at everyone at the interview loop has what I call ‘veto power’, and that’s honestly a very reasonable thing, because nobody wants to hire anyone they don’t want to work with.” ● 把这一轮当成一个随时可以行使、而且有时候会基于很薄的理由行使的否决权。它不是装饰，但它也不是一份评分表。

有一条结构性的事实值得随身带着：创始人轮通常是留给特定人的。“some small companies have this CTO/Founder round especially for senior roles since they want to chat with every person getting hired. Junior employees wont have this round.” ◆ 在大约二十人以下的公司，创始人往往还在亲自做技术评估：“if it’s a very small startup like <20 people then these interviews can still be technical bc C-suite is still somewhat involved in R&D.” ◆

2. 真正决定结果的是什么

3. 直觉问题

创始人公开说这一轮跑的是感觉，而且是白纸黑字、署自己名字说的。

Altman：“There are a few things in the interviewing/negotiating process that you should watch out for because they usually mean that person will not be successful in a startup. A focus on title is an example; a focus on things like ‘how many reports will be in my organization’ is an even worse example. You’ll develop a feel for these sorts of issues very quickly; don’t brush them off. If you have a difficult-to-articulate desire to pass, pass.” ◆（有争议）一位创始人说得更露骨：“When hiring for a startup, it’s often gut-feeling that dictates which of the similarly-qualified candidates is the best fit and I rely on this question to come to my conclusion”，而这里所说的那个问题是一道玩笑题。◆（有争议）“Sunday test”也是以同样的调子被当成筛选手段提出来的：“would you be likely to come into the office on a Sunday because you want to hang out with this person? … Treat your values as articles of faith. Screen candidates for these values and be willing to let an otherwise good candidate go if he is not a cultural fit.” ◆（有争议）一位创始人从业余爱好里读驱动力：“if a candidate tells me they run 10 miles a day as a hobby, that’s a signal of a strong internal drive.” ◆（有争议）另一位把“上班时间能不能接电话”当作决定性依据：“candidates who are unwilling to talk by phone during their current work day. This is a clear sign they are not really looking for their next gig.” ◆（有争议）

候选人这一侧的记录，正好对得上一个未经校准的直觉会产出什么。全票通过之后被否决 ●。因为一项早前几轮已经豁免过的学历条件，在三四分钟内被砍掉 ◆。做完一份多日的带回家作业，十五分钟被终止，没有任何反馈 ◆。另一边，全程只聊餐厅的那一轮给出了 offer ◆，还有一次“离开办公室五分钟后”就到的即时 offer，候选人后来把它读成了最坏的信号 ◆。

直说：这一轮比整个流程里任何其他阶段都更不可预测。技术轮有评分表可以准备。这一轮面对的是一个人，而他自陈的方法是：如果说不清为什么想拒掉你，那就拒掉。

由此得出的实际结论关乎资源分配，不是绝望。在这里有回报的准备工作既便宜又有边界：知道他们在做什么，知道你会做什么，带一样能让他们学到东西的东西，带两个关于在没有结构的情况下工作的故事。这是几个小时的量。超过这个量，额外的准备并不能改变一个凭感觉做出的决定 ◇。不要把这个阶段的拒绝当成关于你能力的诊断信息；这份语料里有三例第一手的砍人，都发生在干净利落的技术轮之后 ●。另外，同时跑比“最后一道闸是评分表”时更多的流程 ◇。

有一条结构性的过滤规则值得知道，因为它跟你完全无关：一位评论者的经验法则是，当公司很小而且创始人本身是技术出身时，创始人亲自做技术评估是合理的；当公司大到已经有一层管理层时，那就是自我意识作祟。“if the place is tiny, and the CEO is technical, it can absolutely make sense. So if you have a large enough company to have a layer of senior managers/directors between the C-level and line managers, you probably don’t want the CTO or CEO interviewing regular IC hires directly.” ◆

4. 提出异议：被邀请，但有范围

这一节之所以存在，是因为一个技术上敏锐的读者，在听完一份研究方案二十分钟之内就会找出它的漏洞；也因为语料显示，这个本能在这一轮里要么是你最强的资产，要么就是终结这场面试的东西，完全取决于它以什么形式交付出来。

本节末尾那三句写好的话是形状，不是台词。里面的细节是编出来的，用来展示异议在问题中的位置，方括号标出的是只有你才能填的部分。在真正说出口之前，用你自己工作里真实的东西把每一处填上。

5. PI 变体

在研究组、研究机构或国家实验室，同一轮依然存在，但重心变了。

技术追问从第一分钟就开始，而且这通电话可以提前结束。一位刚刚跑完三十多场博后面试的 PI：“If you’ve declared proficiency in methods that are the bread and butter of my group you should be able to answer technical questions that only someone who has through deeply about the problems can answer. I started asking these questions at the start of the interview. If during this technical portion of the interview it becomes apparent a candidate has lied about the expertise I end the interview early.” ● 另一位描述同样的探测：“Some postdocs come with oversold skills - e.g. they say they can do something, but it turns out they saw someone else do it, or someone else in the old lab contributed that part to the papers. I try and figure out true skills early as well.” ● 第三位：“I ask technical questions during interviews to evaluate if they understand the techniques and research objectives of my projects.” ●

由此得出的实际结论关乎你自己的简历，不关乎复习。你声称的每一个方法，从最初几分钟起就是一个活靶子，而区分性的问题是：这件事是你做的，还是你看别人做的。

做功课是门槛，不是加分。一位 PI 的底线：“Read or at least skin some of my papers and talk about that here.” ● 另一位把它扩展到日程上的每一个人：“Familiarize yourself with the lab’s work and be ready to delve into discussions about your own research. … Typically, you will receive a list of individuals you will be meeting with well in advance. Take the time to research their work extensively.” ● 一位写博后经验的博主给出的是强版本：“Make sure you’ve read all their recent papers and can discuss your opinions on their research. Come with questions and ideas.” ● 失败被直接点名为红旗：声称对某个列出的研究方向感兴趣，但 “it’s clear they don’t know anything about it.” ●

你被预期带着一个方向来。“Have an idea what you would like to work on in this lab. Like if the PI asks you ok what would you work on in my lab you need to be able to make a simple proposal.” ● 一位 PI 要的是写下来，而且要快：“it’s ideal to have an idea of what you want to be doing (& be able to provide a 2-page research proposal quickly) but be flexible around this.” ● 这份提案可能在第一通电话之后被设成一道闸：“the PI asked me to write a single-page proposal to decide if he would move forward with me.” ◆ 其明说的目的并不是那份文档本身：“they wanted to assess my ability to conceive ideas and the extent to which I can think about science. In terms of format it’s really insignificant.” ◆ 而后续动作和初稿一样重要：“If we are then discussing moving forward together I am looking for how you take feedback, make edits to your research proposal, etc.” ◆

判断错了他们对哪个方向感兴趣，是有记录的被砍掉的原因。对一个带着强烈想法到场的候选人来说，这是那个具体的风险。“Be prepared to talk about how your skillset is going to help me push my research program forward. Don’t be the candidate that only talks about you and how you want to do X so you can do Y. Talk about how you want to do X, so you can do Y within the context of my research program. A top candidate got dropped from second visits because all they talked about was how they wanted to develop their tool for a question that I didn’t think was exceptionally interesting and that didn’t fit into my program at all.” ● 另一位把它列为自己长期的心头刺：“my pet peeve is the postdoc that takes up a position for a project but rather works on his interests or his career without regard for the project.” ● 还有更直白的：“Some postdocs come with preconceived notions that they will do studies they want to do in my lab, usually informed by the graduate mentor. That’s ‘okay,’ but in my lab, we need to do my work.” ●

解法是：带一个方向，它是他们研究计划的延伸，而不是往里面进口一个东西，并且说清楚它挂在他们的哪一条目标上 ◇。

职业目标那道题考的是独立性，不是忠诚度。“What I want to know is people are going to take active control of their career path.” ● “When I (we perhaps) say 5 year plan, I don’t mean a form document. I mean that you, a mature, ambitious, and intelligent adult have a clear, and articulatable goals for your future and that you can convince me that spending 5-10 % of your working adult life in my group is going to benefit those goals. … a graduate student not having goals for grad school is a red flag, but a postdoc without them is a fire alarm.” ● 这里的含糊被直接列进红旗：“a lack of motivation and vision. Vague answers to questions about goals, 5 year plans, and how my group fits into them.” ●

附带两条警告。你说出口的目标会被拿来当过滤器，有时候是过滤掉你：一位 PI 谈到一个瞄准工业界的候选人时说，“I’m not really qualified to advise them about industrial opportunities, so I would probably turn them away. That’s one filter.” ◆ 另外，不同的组对你何时离开的预期差别极大：“my postdoc PI expects us to leave after 4 years to become group leaders. if that’s not our goal, we should do postdoc somewhere else. this is very different from my phd PI who let her postdocs stay for decades.” ◆ 在你主动给出一个可能对不上的答案之前，先问清楚结束的时候会发生什么。

契合度可以压过发表记录。一位 PI 明说的排序：“If they’re absolutely stellar and have the last 4 bullets of the Green list, then great. Otherwise, I will go for people with average CVs who are great to work with over stellar people with potential red flags.” ● 另一位：“This intellectual and interpersonal fit is waaaaay more important than simply attracting the most stellar candidates who graduate with 15 or 20 publications.” ●

这个流程几乎没有任何外部约束。一位正教授：“If these are met sufficiently, and I am confident in the ‘fit’ for a particular project, I will strategically neglect to request references as I won’t want to be talked-out of a candidate.” ◆ 一个人的判断承载整个决定，而且往往没有人力资源部门作为制衡，两个方向上都没有。这是双刃的：一次拒绝可能以一通带着真实反馈的私人电话的形式到来，“they said the interview itself went well, but they went with the senior postdoc candidate who had more experience” ◆；也可能这个决定从来就不会被告知：“A very common experience is that the PI never tells you anything, one way or the other. Many interviews end with ‘let’s stay in touch’ and then they never email you back.” ●

“非正式”这个说法并不让它变得非正式。“Just to note you can’t get around rules on what you can and can’t ask in an interview just by calling it an informal chat. … Having an informal chat is great for screening and getting down to a shortlist, however it very much is part of the recruitment process and both sides need to treat it as such.” ● 一位记录自己求职过程的神经科学家把这通电话叫作它本来的名字：“You can think of this as an informal screening interview.” ●

有兴趣和有钱是两件独立的事。“Some principal investigators may express interest in hiring you upon application but may need time to secure funding for the position.” ◆ 而且口头 offer 不是 offer：一位候选人 “given a verbal job offer and given a tour of the facilities”，然后在研究所的人力资源流程始终没把它兑现之后被失联。◆ 由此引出两条实务。尽早提签证身份，因为 “Most PIs are clueless about international visas so you have to be explicit early on as not to supprise them” ◆，而且机构对国际博后的默认选择常常是 J-1，理由是成本而不是偏好 ◆。另外，预期第一通电话只是一个阶段，不是整个流程：常见的序列是一次三十分钟的通话，然后是一场报告和一整天的一对一 ●。

6. 该问创始人什么

只列有证据支撑的问题。问法按语料里出现的原样给出。

关于现金能撑多久和能不能活下去。问，而且问得简单。一位创始人在回答“这样问是否可以接受”时说：“I like when people ask and I enjoy answering. If it’s important to you, you should ask. … Asking about financials shows that you’re aware there are risks working at a startup and you’re evaluating them, which is exactly what you should be doing and what they should want you to do. On how to ask? - You don’t need to be super analytical about it. Ex. ‘How many weeks of runway do you have in the bank?’” ◆

这位创始人的但书是有记录的那一条：让人不舒服的是过度量化的问法，不是这个话题本身。一位联合创始人按资历划线：“Generally questions regarding stability phrased sensibly will be fine. Asking the exact burn rate is a little too probing if it’s a lower level position. Runway and funding details is fine though, in my opinion.” ◆ 所以：以周或月为单位问还能撑多久，可以。精确的烧钱速度、稀释比例、股权结构细节、毛利率，从一个初级候选人或第一份工业界工作的人嘴里问出来，会读得很糟。注意这块地方是有争议的。一位生物技术圈的评论者说 “Dont ask much about funding or investors or negative structural changes. They dont like that much from my previous talks” ◆（有争议），一位 A 轮之前的创始人说这一整类问题 “so hard to judge as to be hardly worth it” ◆（有争议），还有一位评论者称 “even asking these questions to a first time startup will probably prevent you from getting the job” ◆（有争议）。反方：“You absolutely can, and should. Startup founders have to answer these questions every time they talk to investors; if they don’t have a well reasoned and rehearsed answer they probably aren’t talking to investors much, and that’s a huge red flag.” ◆

不管答案是什么，反应本身就是数据。一位候选人：“I asked this question to the founder, he got VERY defensive and borderline berated me for how dare you ask me these questions …. anyway, I rejected the offer.” ◆

关于风险真正在哪里。来自一份流传很广的候选人清单：“What can we fuck up and this will still be a success” 和 “What can we absolutely not fuck up, because doing so would spell doom for the venture?” ◆ 这两句比问现金能撑多久做的功更多，因为它们要求创始人把那个承重的假设说出口。

关于商业模式。“how does the business make money and who is the primary sales person? If the CEO is still the head of sales, it tells you the startup haven’t yet figured out it’s revenue model.” ◆

关于创始人本人。先看履历：“Ask if the founders have previous startups with successful exits.” ◆ 提出这条的工程师在三家由首次创业者带的公司待过，他把这一条排在现金能撑多久之上。还有直接的版本，来自同一份候选人清单：“What are aspects of your personalities as founder(s) that might make you difficult to work with on a company level? … what are your quirks that we should be prepared to face as an organization?” ◆

关于股权，如果它在谈判桌上。“What % does your equity represent of the total company? If they don’t tell you this directly, then RUN AWAY.” ◆ 以及：“What kind of options does the founder/founders have? You want them to have the same class of options as employees; otherwise, incentives are misaligned from the get-go.” ◆

关于这家机构真正看重什么。一位前招聘人员、现任首席执行官：问“表现最好的那批人有什么共同点”，“gives you insight to what the company values.” ● 他明说这个问题有双重目的，既传递你是认真的，也让你反过来筛他们。

对 PI，另一份清单。来自一位神经科学家自己有记录的博后求职：“What is your philosophy for when a postdoc leaves to start a faculty position? What can I take with me (ideas, transgenic lines, newly developed tech, etc.)? Do you still write grants and publish on the same topics after I leave? … How are conversations surrounding authorship and credit had? … What’s the funding portfolio for the lab look like? … Do you have any plans of moving the lab or changing careers over the next 5 years?” ◆ 在国家实验室，再加上一位评论者认为决定这个位置值不值钱的两条：接收方是会把你吸收掉还是会把你推荐出去，以及 “Will you be able to publish, patent or get anything out that is tangible.” ◆ 工作许可的问题，“straight but subtle and politely with your hiring PI.” ◆

有一个框定方式让上面所有这些问起来都安全：同一位神经科学家说，他问这些问题 “because I genuinely cared for the answers, but also because they served as a litmus indicator for me - to see whether the interviewer respects me enough to engage transparently rather than being offended by very fair (even if unusual) questions.” ●

7. 这一轮告诉你他们是什么样的人

这份语料里的候选人一致报告，创始人轮，而不是任何其他阶段，是他们看清这家机构真实性格的地方。这些迹象很具体。

他们怎么对待不属于他们的信息。一位创始人，对候选人的保密义务：“Those NDAs are of no importance to me. Give me the names.” 当候选人终止了这场面试，创始人的结束语是 “Eat shit and die.” ◆ 这同一位创始人开场时就基于一家第三方猎头提供的数字指控候选人撒谎：“what you just said is a lie, the numbers I have in front of me are saying otherwise.” ◆

他们在没被问的情况下主动说了什么关于工时的话。“I was also thoroughly intimidated by the amount of times he re-iterated that the job will ruin work/life balance and how the ‘easier’ weeks will be 50 hour work weeks. I asked ‘how do breaks work there’ and he replied ‘breaks?’” ◆ 另一例：那位首席执行官 “said that ‘science doesn’t wait for us, so you might have to come in during the weekends’.” ◆ 一位评论者对前一例的判读是对的：“I would take those comments as a major red flag - the 50 hour, ‘less busy’ weeks and acting confused by breaks.” ●

他们在薪酬问题上的行为方式。要求先承诺再披露：“they ask candidates to commit before sharing any details. I told him that didn’t seem like such a great idea, and he assured me that comp wouldn’t be an issue, and that they do this to avoid hiring mercenaries.” ◆ 临门一脚把股权换掉现金：“he lowballed massively and dangled stock options as an alternate to base pay.” ◆ 另外注意，在创始人轮里，薪酬可能就在房间里当场谈，而不是事后由招聘人员来谈 ◆；还有一位科学家出身的首席执行官旁听了候选人和人力资源部门的薪资谈话，候选人觉得这不寻常 ◆。

他们能不能接住一个关于自己领域的硬问题。语料里点名的最清楚的单条红旗：“If they can’t answer tough questions on the science or the business model.” ◆ 它的升级版是一位创始人转而捍卫自己的科学品味：在一处证据双向都有的文献上坚持一边，把 “all of those papers are wrong” 一句带过，然后在参观结束时侮辱了候选人的博后导师。◆

这个房间里进行的是对话还是广播。“Many CEOs are detached from reality, you’ll be listening to stories and anecdotal evidence for his favorite theory.” ◆ 一个相关的迹象是对着错误听众做的路演：“it almost feels as though they are trying to convince me as an investor in the company rather than a worker”，候选人把它读成可能有问题的信号 ◆。反过来的失败也存在：“CEO was complete robot and didnt even try to sell me on the role.” ◆

这位创始人能不能做决定。流程上的混乱被归因到上面，而且归对了：“9 Zoom calls spread out over 3 months was ridiculous, showing a lack of organization and decision making from the company (and ultimately the CEO, who in this case was the hiring manager).” ◆ 另一个极端上，一个“离开办公室五分钟后”就到的 offer 本身，成了一轮本来就已经很奇怪的面试里最大的警告 ◆。

权威是靠论证行使还是靠位阶行使。“In a seed stage startup, the likelihood of a tyrannical CEO bossing everyone around and rejecting process is higher than larger companies. You want to try to catch that right away. A heavily involved CEO is likely, a technical CEO is fine, but a CEO who pulls ‘I’m the boss and I say work on this’ is a red flag.” ◆

面试房间之外的行为。一位候选人在两场面试之间的滑雪道上碰到了一位首席执行官，看着他抢道插到别人前面、不看后面就停下、插缆车队、冲挡路的人喊叫：“It turned out to match his management style 1000%.” ◆ 另一位在面试里当场听到 “but you don’t look autistic!”，他的反思是：“When people show you who they are, believe them and all that.” ◆

一位做过四次这件事的人给出的综合判断。“Runway is important, but just because they have two or 5 or 10 years of runway doesn’t mean they aren’t gonna lay your ass off in 2 months during a ‘pivot’. Every startup is draped in red flags. 9 out of 10 dont make it. … What you want to focus on is (a) ability to build transferrable skills. (b) integrity in the leadership/founders. Do they make outlandish promises (you will be the one burdened with fulfilling them!)? Do they have a track record? Do they have massive egos? Are they going to pressure you to cut corners on regulatory or quality or safety? These are the kind of things you can suss out with careful questioning in interviews.” ●

有一条应该去解读而不是害怕：如果创始人最近刚从首席执行官的位子上退下来，一位生物技术圈评论者的判读是，“founder rarely resigned on their own when they were a CEO. They are almost always being told to ‘resign’ by the board of investors due to various reason (scientific, strategy, etc).” ◆

8. 有争议

以下主张在证据中被标记为有争议，此处不作裁决，并给出每一条被争议的理由。

1. 在大约一百人以下的公司，创始人轮近乎走形式（“he’s not going to overrule the hiring manager unless you totally flub it”）。有争议，因为同一帖子里往下两条回复的那位评论者，正是在这一阶段被否决的。
2. 创始人轮是演戏（“nigh on pointless … they’re not going to block you”）。被同一帖子里的第一手否决案例推翻。
3. 创始人亲自面试在结构上更糟，因为创始人缺乏面试上的成熟度（“the person interviewing me lacked the maturity, experience, or professionalism”）。被那些认为自己的判断就是公司能拿到的最好信号的创始人反驳。
4. 除非被问，否则永远不要提出你对他们业务的看法（“the interview is NEVER a good place and time to discuss your thoughts on their business if not being asked”）。有争议，因为有名有姓的创始人把未经提示的洞见当成他们正在猎取的那个信号本身。
5. 坦率会被惩罚，所以答得更套路一点。基于同样的理由有争议，而且这条建议是在一次从未被告知真实原因的拒绝之后回溯拼凑出来的。
6. 在生物技术公司的创始人轮里，不要问资金、投资人或结构性的负面变化。被同一帖子里其他评论者、以及那些说欢迎这个问题的创始人直接推翻。
7. 对候选人来说，问现金能撑多久、增长计划和产品市场匹配几乎没有用（来自一位 A 轮之前的创始人：“so hard to judge as to be hardly worth it”）。同一帖子里的其他评论者认为这些问题至关重要。
8. 对一位首次创业的创始人提尽调类的问题，很可能让你丢掉这份工作。这是“该问什么”这个话题里最核心的未决争议；证据里既有一位欢迎这些问题的创始人，也有一位因为候选人这么问而斥责他的创始人。
9. 让创始人向你路演，如果这次路演打动不了你，“they are probably losers”。有争议，因为这个启发式奖励的是个人魅力而不是实质。
10. 在公司达到五百人之前，创始人应该面试每一个候选人。这条主张确实存在且有明确出处；但这个具体的人数阈值被广泛认为不切实际。
11. 如果你有一种说不清楚的想拒掉对方的冲动，那就拒掉。被候选人一方反驳，认为这正是未经审视的偏见进入小公司招聘的通道。
12. 把“Sunday test”当作价值观筛选。有争议，因为它筛出来的是与创始人在社交上相似的人。
13. 在几位资历相当的最终候选人之间靠直觉决定，用一道玩笑题。这是“最后一轮跑的是直觉”的最清楚表述；有争议，因为它是非结构化的。
14. 把执着的业余爱好当成内在驱动力的替代指标（每天跑十英里被当成招聘信号）。有争议，因为这个替代指标惩罚了有照护责任或健康限制的人。
15. 不愿意在当前工作日接电话，证明缺乏兴趣。有争议，因为它惩罚了当前雇主管得严的候选人。
16. 略低于市场的薪水加上一份慷慨的股权包，是早期员工的正确结构。作者自己就预料到他会因此 “flamed”；候选人普遍不认同创业公司股权能补偿现金上的折扣。
17. “Any time you have that ‘Wow!’ kind of feeling you need to just hire the person.”与之配套的那项检验，即候选人能否改变你的想法，是有充分支撑的；但挂在它上面的“凭感觉就录”这条规则，因非结构化而有争议。

通话很短，所以谈砸了

错，而且证据在两个方向上都有。一场排了六十分钟的通话在第十九分钟结束，结果是拿到第二轮面试的邀请。一场十一分钟的筛选轮走到了终轮。在某家公司，一位候选人的两场筛选轮里较长的那一场，恰恰是以被拒收场的那一场。●

时长跟着做筛选的人的日程走，也跟着他们多快拿到了需要的答案走。它不跟着他们的判断走。

他们说会把我的简历转给用人经理

这经常是一句写好的收场词，不是一个决定。招聘人员用自己的话描述过这个手法。

一位：“My go-to tactic is the imaginary bad guy. Whenever my prescreens feel like they are obviously not the right candidate I will complete the prescreen as normal and wrap up by saying something about reviewing with the hiring managers.” ◆

另一位：“I would tell them I’ll send their resume to the hiring manager and see what they say. Then reject the candidate a few days later.” ◆

原因与其说是不诚实，不如说是风险管理。一位的说法是，“If you’re rejecting for personality issues, you can’t do that over the phone.” 把拒绝往后推，就把它挪进了书面，在那里可以用安全的措辞写出来。●

他们问我有没有问题，说明谈得不错

有时候恰好相反。两位招聘人员描述过，收尾的提问环节被用来从一场已经判死的通话里脱身。

“As soon as I realize that they’re not a fit, I stop asking questions and say, ‘that’s the last question I have. We’ve still got a minute. Do you have any questions for me?’” ◆

而它对你不可见的原因：“the candidate doesn’t really know what your pre screening questions are, so they aren’t gonna know the difference between you stopping at question 4/10 or question 10/10.” ◆

有一位招聘人员给出了相反方向的说法：等她问到这一句时，她心里已经有数了。这个说法与上面两种读法都兼容。把“走到了收尾提问”当作中性信号处理。●

聊得很融洽，气氛很好

融洽是刻意制造出来的，而且几乎不花成本。一位招聘人员这样描述自己的工作：即使三分钟内就明显看出候选人不合适，这通电话仍然会被拉满到预定的时长。◆

融洽也正是做筛选的人在职业上练过、而你没有练过的那一部分。它是这场通话里最不具诊断力的东西。

我还没有收到回音，应该是出局了

沉默至少有三种不同的结构性来源，其中只有一种意味着被拒。

很多时候是根本不允许说。 “We’re not allowed to give feedback to anyone unless they’re in final rounds for legal protection.” 以及：“Most large companies literally forbid recruiters to give specific feedback.” ●

在某些赛道上，时钟确实以月为单位在走。 在大药企，约到第一通电话要三周，从筛选轮到用人经理十到十四天，都属正常；有一个计算岗位的汇总数字是大约一百二十天。●

有时候沉默是流程在朝对你有利的方向运转。 有一个记录在案的例子：候选人说明自己需要担保，招聘人员就此没了声音，候选人认定自己已被拒。招聘人员实际上在跟用人经理谈判，结果是电话筛选轮被跳过，候选人直接进了用人经理那一轮。◆

他们告诉了我被拒的原因

写出来的理由通常不是真正起作用的那个。一位招聘人员：“We all want to say it’s skillset, but the reality is it’s so often their presentation or conduct.” ◆

一项针对技术招聘人员的研究在简历环节发现了同样的背离：他们给出的淘汰理由，与真正能预测淘汰的因素并不一致。◆

拒绝理由是按照法律上安全、并且不给对方可操作把柄的标准写出来的。不要围绕其中某一条重新设计你的准备。

做筛选的人似乎对我的技术工作没什么反应

在四条赛道中的三条上，这会有意义。在招聘人员这条赛道上没有，因为这个人没有能力评估它。招聘人员对此直言不讳。其中一位给出的有用重述是：筛选轮问的是候选人能不能条理清楚地讲这份工作，用人经理那一轮问的才是这份工作是不是真的。●

一个重要的推论：技术深度用错了轮次，读出来是判断力差，而不是实力强。一段来自候选人一侧的记述：“If you are going deep into system design, requirements gathering, and mentoring during a ten-minute recruiter call, you are likely over-talking the round.” ◆

什么才真正带信息

比人们希望的少得多，而且大多是程序性的，不是情绪性的。

一个具体的下一步，带着指名的日期，或者在通话中、通话刚结束时就发来的排期链接。被问到你在某几个具体时段是否有空，而不是被泛泛问什么时候方便。做筛选的人点出你下一步会见到的具体是哪些人。被要求发来某样具体的东西，一篇论文、一次报告、一份写作样本。这里每一件都代表组织必须付出的工作量，而组织不会为已经丢掉的候选人做工作。◇

诚实的总结是：第一通电话接近一次你从内部读不出结果的抛硬币，而正确的反应是去跑下一份申请，不是回过头来复盘这一份。

之前的三十分钟

六件事，按这个顺序。

认清赛道。 下面所有内容都取决于它。谁约的这通电话，他的头衔是什么，时段有多长，邀请里有没有描述下一步。

准备好与赛道匹配的开场。 面对小型机构或者一个研究组，用研究内容本身，因为对方有能力评估，而且期待你这么做。面对大药企，用一个结构化的行为问题答案，因为对方无法评估科学内容，他是在照着一套胜任力框架打分。面对大型科技公司，用一段紧凑、便于走流程的概述。

准备好一个薪酬数字，并且知道在这条赛道上风险偏向哪一边。 在有薪档的机构里，报得高于薪档会直接结束这通电话。在没有薪档的机构里，报低了会实实在在损失钱。

把你关于担保的那句话写下来并且练熟。 不要临场发挥。这是整通电话里对你代价最高的一句话，临场发挥正是人们要么说过头、要么说出自己撑不住的话的原因。

准备两个问题，哪怕只剩四分钟也要问。 什么都不问是整个语料里被点名次数最多的红旗。

面前放一页纸，上面是你自己的数字。 数据集规模、运行时间、超过了哪些基线以及超过多少、哪些是你本人做的、哪些是整个组做的。在研究岗位的招聘中，把个人贡献区分清楚是一条明写的评分标准，而这正是候选人最常含混过去的地方。

通话中

把提到的所有人名、被描述的下一步、以及跟它绑在一起的任何日期都记下来。这些是你能拿到的仅有的可靠信号，而且很容易丢掉。

当一个回答是抽象的，就去要最近的一个实例，而不是要一段一般性描述。“Can you give me a recent example” 和 “how often does that happen” 是回报最高的两个追问，因为一条没有实例的政策不构成证据。

任何有实质后果的内容，在往下走之前复述一遍。一句话：“So if I am hearing you right, the band is X and the first round after this is Y. Is that right?” 这不花任何成本，却把含糊的说法转成了你日后可以依靠的东西。

如果一个问题是故意含糊的，就把这份模糊性本身当成题目。先问对方指的是什么，然后当场出声给一个粗略答案，并在他面前把它改好。在研究机构里，被衡量的是这个循环的速度，不是最终答案的完成度。

通话后

这一部分的证据很弱，与其塞给你一个模板，不如老实说明这一点。

跟进消息在整个语料里到处都是，几乎没有哪一条记述报告出效果。在研究组这条赛道上，不回复是常态而不是判决：“A very common experience is that the PI never tells you anything, one way or the other. Many interviews end with ‘let’s stay in touch’ and then they never email you back.” 有一家大药企雇主明确要求候选人与自己的招聘专员保持联系，这是整个语料里最接近一份邀请的东西。◆

所以：想发一封简短的信就发，因为它不花成本，而且有一条赛道确实在邀请你这么做。不要从回复或者没有回复里读出任何东西，也不要建立什么跟进策略。◇

通话之后唯一有证据支持的动作是程序性的。如果对方点名了一个具体的下一步和日期，而那个日期过去了，就问一次。如果根本没有人说过下一步，那么你是在通话当中学到了这件事，不是在通话之后。

证据真正奖励的那种状态

两段记述，来自语料的两端，落在同一个不起眼的点上。

一位在糟糕的市场里通过了用人经理筛选轮的候选人：“Smile and look at camera. Talk slower. Don’t over think.”

中文大意：笑一笑，看镜头。说慢一点。别想太多。

一位事先花了几个月把这通电话越想越重的候选人：“honestly I was overthinking it… At the end of the day I had to realize that he’s human just like me and I was just being myself.”

中文大意：老实说我想得太多了……到头来我得承认，他跟我一样是个人，而我不过是在做我自己。

对照本书其余部分的结构性发现，这并不构成矛盾。这一轮里你能控制的部分很窄，你控制不了的部分很宽。把窄的那部分准备透，然后不去表演，这就是全部技术。

分歧登记

报一个高数字会不会让你被过滤掉。 实名的招聘人员说，薪酬问题不是陷阱，它存在是为了让他们能替你争取。匿名的招聘人员说，一个高于区间的数字会终结对话；有一次受控测试中，同一个人以若干个身份申请同一个岗位，只改变薪酬要求这一项，结果是要到上限和高于上限的版本在一两天内被拒，而要最低值的版本每次都拿到面试。归因：利益，外加一处阶段混淆。两边在不同阶段上都成立。过滤和定 offer 的行为方式不一样。

报一个低数字会不会给你的 offer 封顶。 一位大型科技公司的招聘人员说不会，薪档下限仍然适用。在没有薪档的机构以及中介那边的候选人报告说，他们要多少就拿到多少，一分不多。归因：赛道，具体说是有没有正式薪档和内部公平性政策。刚性薪档会把你拉到下限；没有薪档，你报的数字就变成了 offer。另外注意，本语料中没有任何可追溯的案例能证明在有薪档的雇主那里，筛选轮报出的数字给最终 offer 封了顶，所以在那条赛道内部，保护性的说法有支持，损害性的说法没有。

该不该拒绝先报数字。 通行建议是回避，反过来问对方的区间。有一位候选人精确执行了这一套，招聘人员升级到 “I NEED to know what is negotiable now”，然后结束了通话。另一些人报告回避成功。若干雇主要求先给出一个区间才安排后续轮次。归因：政策与执行的差异，外加招聘人员的容忍度。语料里反复出现的停止规则是：回避一次，不要回避第二次。

自我介绍应该多长。 同一个论坛、同一年里的招聘人员，有人说三到四分钟，有人说三十秒以内。有几位说他们根本不问这个问题。归因：没有。提这个问题的人之间本来就没有共识。准备一个三十秒的版本，能在对方要求时展开。

要不要问对方对你有什么顾虑。 招聘人员对开放版本倾向于反感，把它形容成偷懒或者急切，但并不一致。用人经理倾向于欢迎，同样并不一致。归因：两边说的是不同种类的话。每一位报告了真实实例的用人经理都说它有效，并描述自己因此回头去谈一个真实的差距。两边的反对意见都是关于套路化问题的政策偏好，而不是它失败的记录。剩下的由措辞决定：指向某件已经谈过的具体事情的窄版本，表现优于开口索要一个判决的开放版本，而开放版本有一例记录在案的反效果。

什么时候说明自己需要担保。 候选人报告在招聘人员这一环节被当场拒掉，也报告在一家大雇主那里正是靠着把这个信息压到用人经理那一轮才走到 offer。招聘人员把隐瞒称作一面橙旗。归因：招聘人员一侧是利益，候选人一侧是真实代价。未解决，而且两边的代价都有记录。

cap-exempt 身份是优势还是陷阱。 两种立场都出现过，都有亲历者的论证。归因：你说的是哪一种签证。就工作签证本身而言，cap-exempt 身份是实打实的优势，因为它去掉了抽签。就永久居留而言它可能是陷阱，因为这个身份属于雇主，日后转到普通雇主时带不走，而这类机构启动担保的时间又常常晚到走不完流程。两个阵营描述的是同一条路上的不同阶段。

博后走交流访问身份还是走工作签证。 在同一场讨论内部被两个方向都论证过。大学偏好更便宜的交流访问路线，它还能覆盖整个合同期；候选人指向可能附着其上的回国居留要求。未解决。它取决于该要求是否适用于你的国籍和资助来源，这是一个该问移民律师的问题，不是一堆论坛帖子能回答的。

大型科技公司认不认杰出人才签证。 一段记述称它被例行接受，另一段记述称它被拒绝。未解决，而且两边大概都已过时。当作一个要直接去问的问题。

候选人什么时候可以提薪酬和福利。 一个有经验的社区说第一通电话就可以。一位从业三十年的招聘人员和一位大型科技公司的前招聘人员都说要等。未解决。

不对应真实空缺的岗位发布常不常见。 一家供应商说大约五分之一。一家政府研究机构说没有官方统计，并指出发布这些数字的供应商本身就在卖招聘服务。企业招聘人员否认自己这么做。归因：利益。可核实的那一部分是长期挂着的岗位编制，按供应商自己的文档，它本身并不能直接录用你。

走量还是逐份定制。 一位在两家大型科技公司做过招聘的前招聘人员建议高投递量并且尽早投，并描述自己看一份简历不到三十秒，每个岗位大约只看前两百份。对立阵营认为，投到匹配度不高的岗位是浪费。归因：赛道。两个阵营都同意，在大雇主那里做深度定制回报很差。

创始人那一轮是不是走过场。 同一批讨论里的亲历记述，既描述过盖章放行，也描述过在面试小组一致通过之后被否决。归因：没有。这一轮确实没有校准，而这本身就是结论。

在初创公司问财务尽调类问题是加分还是减分。 同一个帖子里的两条评论立场相反。一位尚未完成首轮大额融资的创始人认为这类尽调问题几乎不值一问；同一帖里的一位兼职首席技术官则认为，答不上来是一个严重的警示信号。归因：层级。同样是精确的现金消耗速度问题，从一位资历浅的候选人口中问出来，和从一位资深候选人口中问出来，读起来不一样。

国家实验室在担保上有多宽松。 一段记述报告在 cap-exempt 身份下两周拿到签证。另一段来自在其中一个实验室有直接经验的人，说他们默认走学生和交流访问路线，并点名三个实验室比另外三个更宽松。未解决，而且各实验室之间有差别。

口音是不是一个真实的障碍。 在同一个社区内部，同一个帖子里既有一段记述说自己因为某位总监对其英语口语不满意而被淘汰，也有一段记述说不必担心，因为在那里工作多年的人英语也说得不完美。没有共识。留作开放。

一个并不是分歧的分歧

中文平台上的建议和英文平台上的建议，在薪酬、在诚实、在人力资源那一轮处于流程什么位置这三件事上分歧很大。在一个市场里，那一轮是前置的过滤；在另一个市场里，它排在最后，而且常常只是宣布一个早已在人头编制层面做出的决定。在一个市场里谈判是被预期的；在另一个市场里，开价超出预算会让申请被直接丢掉，连讨论都没有。

这些不是关于同一个系统的互相竞争的主张。它们是对两个不同系统的准确描述。中文平台的材料在本书中只在涉及移民身份的地方保留，因为那部分可以迁移，其余部分一律标注为不可迁移。

证据根本不存在的地方

对研究岗位重要的三个问题，在本语料里完全没有亲历覆盖：算力怎么分配、发表和署名政策是什么、什么东西会开源。不是覆盖稀薄，是没有。

这一点值得明说而不是糊过去，因为已有的问题库本来就把这一族问题标注为从学术场景迁移过来、而非在工业界得到实证，本次收集确认了那个标注是对的。这些问题想要答案就该问，而你也应该问，但要明白没有人写下过关于它们落地效果的证据，本书不假装不是这样。

关于计数的一点说明

许多帖子表面上的一致，未必是独立的一致。有一条写于 2024 年的记述，被发现在 2026 年由另一个账号在同一平台上逐字重发。流行的解释会被抄，所以一条出现很多次的说法仍然可能只靠着一个来源。本书中的证据角标反映的是独立来源数，不是出现次数。

你的申请分布在哪里

在你投过的岗位里，四条赛道上的分布大致是这样。小型研究机构，也就是 AI for science 初创公司、小型生物技术公司和非营利研究所，是最大的一组。大药企是第二大的一组，而且它们不是大型科技公司的一个变体，它们从头到尾跑的是另一套流程。研究机构、国家实验室和博后岗位是较小但确实存在的一组。大型科技公司是最小的一组。

实际的后果是：通行的第一通电话建议描述的是一场结构化的招聘人员筛选轮，而它适用的是你四组里最小的那一组。在你最大的那一组里，第一通电话经常由一位科学家来跑，而且立刻进入技术内容。给所有这些准备同一套脚本，等于在为你投得最少的那条赛道做准备。

转专业问题，四条赛道上都会被问到

你从工程力学走到计算生物物理，再走到面向分子结构的机器学习。这条路径足够不寻常，一定会被问到，而证据表明这个回答是按连贯性打分的，不是按单个步骤打分的。

有两条发现直接适用。第一，发表能让你拿到这通电话，然后就不再计分了；深度是在这之后被检验的，而发表记录漂亮的候选人被追问时经常被发现有空洞。第二，在研究岗位的招聘中，把个人贡献区分清楚是一条明写的标准。你被期待主动说清楚哪些部分是你做的、哪些是别人做的，不需要人问第二遍。

所以这个转专业的回答应该是一句连贯性，后面接证据，而不是一段编年史。在你的情况里这条连贯性是真实的：力学背景提供了物理和数值方法，博士工作把它们用到分子上。这句话说一次，然后立刻转到你造过的某样具体东西以及它做到了什么。不要复述学位的先后顺序。

担保

你需要它，而它是第一通电话里对你代价最高的单一话题。有三条发现塑造了处理方式。

它常常在任何人跟你说话之前就已经定了。在大药企，人力资源会在用人经理看到你的材料之前就先筛这一项，而那个决定可以推翻用人经理。

主动交代没有被问到的细节，终结过流程。有一位候选人被问了一个关于顾虑的开放问题，提到了自己工作许可剩余的时间；招聘人员把这件事在内部上报，材料再没有被往下传。

回答得不准确同样终结过流程，只是发生得更晚、代价更大，其中一例是在设备已经寄到候选人家里之后。

你已有的两条规则干净地解决了这件事，而且它们本来就是对的规则。绝不说任何不真实的话。绝不主动交代没有被问到的弱点。用到这里：准确回答被问到的那个问题，一句话，然后停。除非被问，不要附上时间线、类别或者复杂情况。这句话要在通话之前写下来，而不是临场发挥。

对一个常见假设的一点纠正。非营利研究所并不自动是安全选项。证据里描述的是，在研究经费削减之后，非营利机构的 cap-exempt 名额已接近拿不到；cap-exempt 身份属于雇主而不属于你；日后转到普通雇主又会把你重新放回抽签里。在国家实验室之间，各方记述描述有些明显比另一些更愿意做担保。

你已经投出去的申请

两个具体问题，都是往后可以修正、无法回溯修正的。

多个邮箱地址。 你用了不止一个地址投递。这件事的效果和它看上去的不一样。申请追踪系统（ATS）按姓名加电话号码加个人主页链接去重，或者按姓名加城市加国家去重，至少有一家根本不看邮箱。所以第二个身份什么也藏不住。它真正造成的，是一批日后会被某位管理员合并的记录，而那个人无法选择哪个版本留下来。从现在起用一个地址、一个电话号码、一个主页链接。

在某家大药企投的那一批不连贯。 你在那里投了七个头衔，从博后科学家一直到总监，横跨 RNA 基因组学、agent 研究、计算化学、人类基因组学和软件开发。证据一致指出，警示信号是不连贯而不是数量，而一批横跨资历层级和不相关职能的申请，正是招聘人员点名的那种模式。有支持的规则是：在一个职能族、一个资历带以内，投两到五个岗位。一位博士候选人去投总监级别的岗位，本身也是一处层级错配。

如果那家公司的招聘人员联系你，能用的最有用的一句话，是不等对方问就告诉他你还投了哪些岗位、以及你真正想要的是哪一个。这把一片乱枪打成了一个明确的偏好，也消除了招聘人员专门描述过的那种协调失灵。

你最强的本能在哪一轮里是负债

你能很快、很准地找出一份研究计划里的弱点。在四条赛道中的三条上，这是资产，而且正是被检验的东西。在创始人或者 PI 那一轮里，它是有条件的资产，而条件很严。

证据真正支持的范围比通行建议要窄，这个差别很重要。创始人欣赏那些看到了他们没看到的东西的候选人，他们也邀请你批评你自己过去的工作、批评你上一个雇主、批评他们自己提出的反对意见。本语料里没有任何一位创始人邀请你批评他当前的计划。与此同时，主动对业务或者核心路线给出判决的候选人因此被拒。在一家小型研究机构里，核心赌注就是科学路线本身，而那恰好是你最擅长找漏洞的东西。

安全的形式是一个关于方法的问题，不是一个关于前提的判决，而且最好在被邀请之后再提。具体说，差别在于：说一个路线行不通，和问他们打算怎么处理那件会让它行不通的具体事情。第二句话包含同样的信息，展示同样的判断力，而且不要求对面的人去为自己创办这家公司的决定辩护。

三句这种形状的话，用你自己的领域：

“Can I ask how you build the test split? I ask because the thing I would want to check first in my own work is whether close homologs end up on both sides, and I have been burned by that.”

“How long is the loop between a set of predictions going out and the experimental results coming back? I am trying to understand whether the model gets retrained inside a campaign or whether the design is effectively frozen between rounds.”

“When you quote that accuracy number, is it measured against your own assay data or against the public benchmark? Those two have come apart badly in the systems I have worked on, and it changes what I would prioritize in the first six months.”

每一句都点出一个真实的失效模式，把它挂在你自己的经历上而不是挂在对方的能力上，并且给对方留出回答的余地，而不是辩护的余地。

在你更早的准备里，有人建议你在面试中对一家公司公开的技术方案提出一个结构性的反对。那条建议在实质上是对的，在表达方式上是错的。把它作为一个问题提出来，范围限定在方法上，并且最好在被问到你的看法之后。

你手上唯一的直接数据点

你在一家非营利研究所面试过，之后被拒。那是你在研究机构这条赛道上唯一属于你自己的记述，它比本书里任何一份外部报告都更有价值。如果你还留着当时的笔记，把它们对照研究机构那一节重读一遍：第一次谈话是不是立刻进入了技术内容，你是不是被期待带着一个方向来，以及有没有人力资源的人出现在科学内容之前。