1. 触达 → 判定：交「他自己说过的一句具体现状」，不是「他好像有兴趣」。原话记下来，后面每一格都要用它。
2. 判定 → 发现：交「他符合理想客户画像的哪一条，凭什么」。写不出这一条，说明这单是靠感觉进来的。
3. 发现 → 决策链：交「一句他能自己复述的痛，且已经换算成钱或时间」。他复述不出来，就等于没有。
4. 决策链 → 方案：交一份名单：谁掏钱、谁使用、谁会反对，每人一句他在意什么。
5. 方案 → 成交：交一个具体的下一步——谁、做什么、哪天。「我们保持联系」不是下一步。
做法：这五句话写在同一处（CRM 备注就行），每次动作之前先读一遍。哪一句是空的，就回去补，别往下走——往下走只是把空缺带进更贵的格子。
EN You debrief by asking which step you botched, but six stages at 80% each don't add up to 80% end to end. Five handoffs losing a fifth apiece leaves you a third of what went in; five handoffs losing a twentieth leave you three quarters. The variable that decides deals isn't performance inside any stage — it's whether each stage hands something specific to the next one. Loss at the seam is the default state, not an accident: working memory holds only about four chunks at once, so the customer detail you were sure you'd remember has been compressed into a vague impression by the next call two weeks later. Vogel and Machizawa (2004, Nature) measured that ceiling directly with contralateral delay activity, an EEG signature that saturates around three or four items and localises near the intraparietal sulcus; Cowan (2001, Behavioral and Brain Sciences) reached the same limit from behavioural evidence. Writing the handoff down as one sentence simply routes around the bottleneck. There's a large-sample precedent for structuring the seam: Starmer and colleagues (2014, New England Journal of Medicine) introduced the I-PASS handoff bundle across nine children's hospitals covering 10,740 admissions and saw medical errors fall from 24.5 to 18.8 per 100 admissions and preventable adverse events from 4.7 to 3.3, with no increase in handoff duration. That's a pre-post study in medicine, not a randomised trial in sales, so treat it as the mechanism rather than the number. So carry five things: the specific line the buyer said about their current state; which criterion of your ideal customer profile they meet and on what evidence; one pain they can repeat back, already converted into money or hours; a list of who pays, who uses and who objects, one line each on what they care about; and a next step naming a person, an action and a date. Keep the five in one place, read them before every move, and refuse to advance a deal on a blank line.
机制：每一格的输入就是上一格的输出。信息在接口处丢失是默认状态而不是意外——人脑一次只能稳住三四个组块，你以为「记住了」的客户细节，两周后已经被压缩成一个模糊印象。把交接物写成一句话外化出去，等于绕开了这个瓶颈。而损耗之所以致命，是因为它相乘不相加：五个 0.8 连乘只剩 0.33。
边界：一次通话就能定的小额生意别套五次交接，那是给自己加工序。这套的价值随周期长度和参与人数上升——两周内、一个人拍板的单子，一句备注就够。
停最久那一格的卡点几乎总是同一件事：对方愿意帮你在内部转述。而转述必然衰减——他不会用你的话说，也不会记得那三个数字。你要做的不是给他更好的材料，是把自己放进那个房间。
可以照念：「我们聊得挺顺，但我不想让你替我在内部转述——通常转述完会掉一半，掉的还偏偏是财务那一半。下周能不能约 30 分钟，把预算这条线上的人拉进来？我准备一页纸，只讲三件事：现在这套一年花掉多少、换过来能省多少、多久看得到。你不用替我讲，你只需要在场说一句你怎么看。」
为什么管用：它让对方少做事而不是多做事，给了具体时长和议程。他要拒绝就得给个理由——而那个理由正是你最缺的信息：要么他并不真想推进，要么预算那条线上另有其人。
EN Here is one deal walked end to end. It's a composite built from common shapes, not a real company's post-mortem; the numbers illustrate proportions and shouldn't be taken as benchmarks. Day 0, a three-sentence cold email built on one specific observation. Day 6, a twenty-minute call that qualifies — budget and timing confirmed, or the deal is killed. Day 19, forty-five minutes of questions where the pain gets converted into roughly 380k a year. Day 34, the person who signs is in the room and an internal champion exists. Day 61, a one-page business case the buyer can repeat back without you. Day 88, a close that yields a go-live date rather than "let us think about it". Month 14, renewal and expansion. Watch the exit conditions, not the lines. The stage that took longest was the decision chain, fifteen days, and that is where most deals stall — almost always for the same reason: your contact offers to take it inside for you. Relaying always attenuates, and the half that gets lost is the financial half. So don't send better material with them; get yourself into the room. Try this: "I don't want you relaying this internally on my behalf — half of it usually gets lost, and it's the finance half. Can we get thirty minutes next week with whoever owns the budget? I'll bring one page covering three things: what the current setup costs per year, what switching saves, and how fast it shows. You don't have to present it. You just have to be there and say what you think." It asks the buyer to do less, not more, names a duration and an agenda, and if they refuse, the reason they give is itself the information you most needed. Underneath all of this sits one finding: a next step that names a person, an action and a date is not politeness, it's the mechanism that makes the step happen.
机制：单子在两格之间停住，通常不是对方不想推进，而是没人对「下一件事」负责到具体的人和具体的日子。把每一格的结束动作固定成「谁、做什么、哪天」，等于把一个需要靠意志力发起的意图，换成一个由情境线索自动触发的动作——这是行为科学里效应最稳的一类干预。
边界：七格是给周期以月计、多人参与的生意用的。个人消费、一个人拍板的场景压到三格就够：判定 → 讲清价值 → 要一个决定。硬套七格只会把自己拖死，也会让对方觉得你在走流程而不是在帮他。
「王总，合同快到期了，续约的事咱们什么时候聊一下？」
问题：这句话把议题设成了「你要不要再付一次钱」，等于请对方从零开始重新评估你。而他手上没有证据，只有印象——印象里最容易被调出来的，正是上线时最难受的那两周。
「王总，还有两个月到期。我先自己做了一版这一年的账：上线第 6 周你们第一次跑通了整条流程，到现在每月大概省下 40 小时——这个数是从你们后台导出来的，不是我拍的，附件里有明细。我想听你说两件事：这一年最不顺的是哪一段，以及明年你最想解决的是什么。下周三下午方便吗？」
差别在哪：你先把「峰」换成一个可核对的事实（第 6 周跑通），再主动把最难受那一段摆上桌——讲出来的痛苦比他自己回忆的版本温和得多。最后一句还把话题从「续不续」推到了「明年解决什么」。
1. 把「首次价值」定成一个可核对的事件。不是上线，是他第一次真的拿到结果。写进交付计划的第一行，并且约定到那天你们一起看一次数。
2. 最难受的那一段，提前说它会难受、大概多久。数据迁移、换流程、内部培训——预期提前抬平，峰值就不记在你头上；事后解释只会被当成找借口。
3. 结尾别用「项目结项」，用一次把成果念出来的会。三个数字，加一句他自己说过的话。二十分钟就够，却是他一年里记得最清楚的二十分钟。
4. 成交那一刻，别多承诺一句。你多说的那一句不会让他更想签，只会把之后 12 个月的比较基线抬高一格。
EN Whether a customer renews doesn't track the average of their year with you. It tracks the peak — the worst or best moment — plus how the last stretch felt. That's why one extra promise at signing costs so much: it doesn't buy you the deal, it raises the baseline against which twelve months of delivery gets compared, and what the brain records is the difference between what happened and what was expected. The evidence here is unusually hard. Kahneman, Fredrickson, Schreiber and Redelmeier (1993, Psychological Science) found people preferred to repeat a longer unpleasant episode with a gentler ending over a shorter one that ended at its worst; Redelmeier and Kahneman (1996, Pain) reproduced it in colonoscopy, where extending the procedure with a less painful final minute lowered remembered pain; and Redelmeier, Katz and Kahneman (2003, Pain) ran the randomised follow-up and found the improved-ending group was more likely to return for a repeat procedure — which is about as close to direct evidence about repeat purchase as this literature gets. Underneath sits the reward machinery: Schultz, Dayan and Montague (1997, Science) showed midbrain dopamine neurons fire to the prediction error, not the absolute outcome — better than expected fires, exactly as expected barely moves, worse is suppressed. So four moves. Define first value as a checkable event rather than go-live, and put it on line one of the delivery plan. Warn them in advance about the ugly stretch and roughly how long it lasts, because an expectation flattened beforehand doesn't get charged to you, while an explanation afterwards reads as an excuse. Replace project sign-off with twenty minutes reading the results out loud — three numbers and one sentence they said themselves. And at signing, say one thing less than you want to. One boundary: peak-end holds most reliably for short, clearly bounded experiences; stretching it over a months-long delivery shared by several people is a reasonable extrapolation, not a demonstrated result, and it is never a licence to let the middle rot — a bad middle just manufactures its own negative peak.
机制：大脑评估一段经历不靠积分，靠取样——峰值和结尾权重极高，持续多久几乎不计（「持续时间忽视」）。同时奖赏系统编码的是预测误差而不是绝对值：好于预期才放电，等于预期几乎不动，差于预期被抑制。两条叠起来的结论是，交付的价值感是差值造成的；所以过度承诺是双份代价——当下不多卖出什么，之后每一次交付都被推进零或负的区间。
边界：别把峰终当成「中间可以随便烂」的许可——中间烂会自己长出一个负峰值，那正是记忆最爱抓的东西。而且这套只在你确实交付了价值时才成立：用体验设计盖住产品没做到的事，最多骗过一个续约周期。
第一步（不许看结果）：七格逐行写「出口条件满足了吗：是 / 否 / 说不清」。「说不清」按「否」算——你当时没有证据。
第二步：找出第一个「否」，写它发生的日期。大多数单子的真实死亡时间，比你以为的早一个月。
第三步（这时才看结果）：把这一单落进上面四格，按那一格的指示行动。
第四步 · 三个问题：①下次在同一格，我可以提前多问哪一个问题？②这一单里有哪件事是我控制不了的，别记在自己账上？③如果这单一开始就注定丢，最早能在哪一格看出来？
关键的一条：复盘要在写下预测之后做才有意义。单子进入报价阶段时先写一行「30 天内成交概率 __%，因为 ____」，复盘时把它翻出来对——否则你只会读到一个「我当时就觉得」的故事。
EN Debrief by outcome and what you learn is the shape of your luck. Go stage by stage first — was the exit condition met, yes, no, or can't tell, and "can't tell" counts as no, because you had no evidence at the time. Then find the first no and date it: most deals actually died a month earlier than people think. Only then bring in the result, and place the deal in one of four boxes. Process clean and won: copy the sequence, not the clever lines. Process clean and lost: don't change your playbook, change your qualification bar — either this deal shouldn't have entered the pipeline or it was luck. Process broken and lost: the one box people actually learn from, and the one where they learn the surface, writing "they thought it was too expensive" instead of naming the stage whose exit condition failed. Process broken and won: the dangerous box, because winning locks in the steps you skipped. The biases here are well-established. Fischhoff (1975, JEP: Human Perception and Performance) demonstrated hindsight bias — once you know the outcome, you systematically overestimate how predictable it was; Guilbault and colleagues (2004) confirmed it meta-analytically across more than a hundred studies. Baron and Hershey (1988, JPSP) showed outcome bias: the identical decision gets rated as better or worse depending only on how it turned out, which is exactly why the order of the two questions matters. Structured debriefs do work when they're actually structured — Tannenbaum and Cerasoli (2013, Human Factors) meta-analysed 46 samples and found debriefs improved performance over controls by roughly a quarter, d ≈ 0.67, with the effect concentrated in debriefs that were structured, focused on specific behaviours and conducted without blame. And none of it works retrospectively: write "probability of closing within 30 days: __%, because ____" when the deal enters the quote stage, then check it. Without a written prediction, a debrief just reads back a story you have already rewritten.
机制：知道结果之后，大脑会自动重写「当初我怎么想的」，并把已发生的结局看成本来就明显。所以按结果起头的复盘，学到的必然是被结果污染的版本。先判过程再看结果，是在结果污染进来之前，把当时的判断锁住。「说不清」按「否」算，防的是同一种自我美化。
边界：复盘有成本，别每单都做。挑三类：金额最大的、周期最长的、和你觉得「莫名其妙就没了」的那种——第三类信息量最高。「不追责」不是场面话，它是效应存在的条件：复盘一旦变成定责会，人只会交出能自保的版本。
90 分钟，把一个真实的单子从头到尾走一遍。