Chapter 7The Clarity Method: Expand Your Options

1. Where the first version of a standard comes from #

Chapter 6 left the team a pause rule. When a piece of work is about to be repeated, expanded, or handed to other people and to AI, and the trade-off and the acceptance still cannot be stated, do not keep raising the volume of execution. The team should turn to finding, borrowing, or verifying a standard.

The pause creates the next problem at once. Suppose a company is about to rebuild its support system, and three groups argue for a larger knowledge base, a redesigned handling flow, and a new AI. Each plan can produce a demo, and each can list a string of benefits. Pushing for more versions at that point only multiplies the options. What may be missing is not a fourth plan. It is the basis for judging the first three: what this rebuild should improve most, which results cannot be sacrificed, and what evidence would be enough to say it works.

The first version of a standard may not come from internal debate. The team may never have handled this kind of problem. The organization's existing experience may also come from another business, another set of users, or another stretch of market. Without outside evidence, seniority can decide on its own. Without a prior record of what information would change the choice, collecting material also has no clear stop.

The Clarity Method is built for that problem. It splits the resource used to solve a problem into three parts: refining a standard (10), collecting data (10), and seeking a better solution by means (80).

Those numbers are not an organizational average from a survey, and they do not require every project to log ten hours, ten hours, and eighty hours. They set an order of magnitude: use a small share of resource first to confirm what the judgment will rest on; then collect the key facts that could change the decision; leave most of the resource for proposing methods, verifying cheaply, comparing results, and continuing to revise. Safety, medicine, regulation, or irreversible decisions will of course need more evidence up front. 10/10/80 also cannot replace professional duty and the evidence a field already requires.

Standards and data are not "worth" only two-tenths. Without a pass line, later tries cannot be compared. Without relevant facts, a borrowed standard cannot be judged for the problem at hand. Resource discipline is there to stop two kinds of waste. One is arguing, in a conference room, over a perfect standard that has not yet met reality. The other is keeping the files coming without ever saying which new piece of information would change the choice.

2. Look first at whether the gap is the angle, the scale, or the conditions of use #

"No standard" is not always a missing score or threshold. Many problems first lack an angle of observation.

Take the support system again. If the team looks only from cost, it will compare cost per ticket and headcount. From the user's side, it also has to watch whether the problem was actually solved, and whether the user had to come back. From risk, it has to check account safety, mistaken promises, and escalation to a person. Cost, user result, and risk are three different dimensions. A dimension decides from which angle the same work is seen.

An angle still needs a way to judge. On user result, the team may require that first-contact resolution for the same class of problem not fall. On risk, an account issue with thin facts may have to go to a person. Ranking, a pass line, and a stop line of that kind are the standard. A standard gives an angle a scale, so that different plans can take the same check.

The scale cannot be detached from the conditions of use. The same auto-handling rule may be reasonable for a low-amount, reversible ordinary inquiry, and wholly inapplicable for medical advice, a large refund, or a case involving a minor. Object, stage, resources, rules, the time window, and the speed of feedback all affect whether a standard is reliable and whether it applies.

Finding a gap in standards, then, often means an important angle was not looked at, and that is not the only case. Sometimes the angle is already known and there is still no scale for comparison. Sometimes the scale exists and the team still cannot tell whether it can be moved to the present scene. When a standard is borrowed, three pieces of information have to come together: which angle it is looking from, what scale it uses to judge, and under what conditions it holds.

Three kinds of material are easy to mix, and they are not the same thing. Physical laws and engineering constraints describe what cannot be violated. Market demand and behavioral data record how users choose under given conditions. An organizational standard says how the company will trade off on the basis of those facts. Physical constraints and market facts can help an organization form a standard. They still are not the standard. That people in the market buy a certain class of product is a fact. "Enter only categories where demand is already verified, and where we can make a difference" is a standard of choice.

3. The core is borrowing, but what you borrow has to be complete #

Many operating problems have shown up somewhere before. Peers may have handled nearby users. Other industries may have solved similar constraints. Regulation, research, and public accidents may also have left a basis of judgment that can be checked. Looking first among those materials for a candidate standard is usually more reliable than paying the whole trial-and-error cost again.

Borrowing does not mean someone else has already produced your answer, and a famous company's method is not a reason to import the whole thing. The value of borrowing is that the team begins judging from understanding other people have already verified, and so repeats less of the same search: which dimensions they saw, what scale they used, and under what conditions they got what result. A borrowed standard is still only a candidate. Present data may show that it does not apply.

The method can be run in five steps:

Find the gap → Borrow a candidate standard and its conditions of use → Check migration → Verify cheaply → Rank by value and write into the standards cycle

When the gap is found, say first whether what is missing is the angle, the scale, or the conditions. When borrowing, do not lift only the slogan a winner finally spoke. Look for the object they faced at the time, the resources, the constraints, the feedback, and the counterexamples. The migration check then has to answer: was the original problem the same one; will the key objects respond in a similar way; and can error show itself inside a cost and a time the organization can bear.

The share of resource used to collect data mainly serves that check. The team does not need to gather every available fact first. It should look first for facts that could overturn the migration: what has already changed about the users, where the cost structure differs, whether regulation has added a new boundary, and which failures the original case did not cover. What no fact could overturn cannot stand as a standard that can be tested.

After it passes a first check, the team still should not copy at full scale. The first action should shrink the range, the budget, and the people affected, and write the evidence of a pass and the stop condition in advance. If the result supports the candidate standard, expand a little. If it does not, separate whether the standard itself was wrong, a condition of use was missed, or execution never happened as agreed.

When there is no object to borrow, or the migration check does not pass, the work should enter exploration. Exploration may leave the answer unknown. It still has to bound the problem, the cost, the evidence, and the stop line. The core of the Clarity Method is borrowing, and that still leaves room for original work. The organization first has to tell whether the problem truly has no precedent, or whether existing experience has not yet been found and written down.

Whether an outside practice can actually help a firm cannot be judged from success stories alone. Miriam Bruhn, Dean Karlan, and Antoinette Schoar ran a randomized controlled trial in the Mexican state of Puebla. Of 432 micro, small, and medium enterprises that wanted to take part, 150 were randomly offered a year of subsidized consulting, and 282 were controls; 80 of the invited firms actually participated (Bruhn, Karlan & Schoar, 2018).

The program did not send the same template to every firm. The firm first diagnosed the problem with a consultant, then was matched, by industry, size, and need, to one of nine local consultancies. For the year that followed, owners worked with the consultant about four hours a week. What was improved differed by firm. Marketing, formal financial records, and longer-term planning were among the more common items. The process sits closer to "borrow a candidate standard, check the present problem, and support execution" than to copying a best-practice list.

After one year, in the main estimates, productivity in the invited group was 0.21 standard deviations higher than in the control group, and return on assets was 0.22 standard deviations higher. Short-term sales, profits, and employment showed no significant growth. Pooled estimates from administrative data over the next five years put employment in the invited group about 57% higher than in the control group, and the wage bill about 72% higher, though the year-by-year estimates are imprecise. The results show that outside management knowledge, after diagnosis and continuing implementation support, can improve operating results for some small and medium firms. The trial does not identify which piece of advice did the work, and it does not test 10/10/80.

Figure: 10/10/80 marks where the weight of resource goes, not a precise ratio; winners and losers help form and test a migration hypothesis, and one pair of cases cannot prove cause.

4. Duan Yongping: let the market prove demand first #

Duan Yongping compressed that starting point into "dare to be last." The publicly compiled notes of a 2018 conversation at Stanford confine "last" to the product category: market demand is hard to guess; after others have already made the demand clear, meeting it carries more certainty. Later-but-better then handles the other half of the problem: a later entrant still needs the ability to make the product well (Duan Yongping, 2018, public compilation).

The remark holds two layers of judgment. The first uses choices users have already made to confirm demand, rather than a forecast inside the company. The second checks whether the company can make a better product inside an existing category. Proof that people in the market want something does not, by itself, prove that a later entrant can win. Evidence of demand removes some of the uncertainty of direction. Product, channel, cost, and organizational capability still have to be built by the later firm.

BBK's VCD is a concrete case. A company chronology held by CCTV shows that BBK had already launched VCD products in 1997: in August it brought in Jet Li to shoot an advertisement, and in December the AB001 was rated a superior product in a quality grading organized by the Ministry of Electronics Industry. It entered a crowded market, not a blank category that still needed an explanation of what a video disc player was for. The company put its resource into quality, service, and brand (CCTV, BBK company chronology archive).

Those facts can confirm that BBK later entered a category where demand already existed, and that a product launch and a quality rating were left on the record. They cannot prove that "entering later" produced the result on its own. Channel, advertising, supply chain, quality, and the competition of the time all sat there together. Duan's public remarks can confirm his method of choice. They are not an experiment that isolates cause.

A January 2025 transcript of a conversation at Zhejiang University adds where the method applies. The early emphasis on daring to be last was related to the company's then-small scale. Once an organization reaches a certain size, and no one sits ahead of it, it also has to do new things. The transcript uses search and phones to make the same point: a later mover can catch up only if it has the strength to make a better product, and even a large company can fail as a latecomer (Duan Yongping, 2025, Zhejiang University transcript).

That boundary brings "dare to be last" closer to a usable standard: demand has already shown itself in real choices; the category still has long-term value; the later firm can make a differentiated product; and the market opportunity has not already been held for the long term by the first movers. Missing any one of those, "someone else has already done it" is only a piece of market information. It is not a reason to enter.

5. Musk: when no ready standard exists, re-derive from physical constraints #

If no one sits ahead, or an existing practice cannot migrate, the team cannot force a source into place for the sake of insisting on "borrow." When Musk talks about first principles, what he shows is a way to handle the case in which no ready standard can be borrowed.

In a 2008 oral history for The Henry Ford, he explained first principles as constructing conclusions upward from well-tested basic facts of physics, rather than being bound by convention and by past analogy. In a 2013 TED interview he put it again as: break the problem to the most certain basic facts, re-derive from there, and actively look for negative feedback (Musk, 2008, 2013).

He gave a concrete example in a 2012 WIRED interview. By his estimate, the raw materials used in a rocket—aluminum, titanium, copper, carbon fiber—are about 2% of a typical selling price. That gap does not prove that the finished product should sell at the price of the raw materials. Design, manufacturing, testing, capital, safety, and the risk of failure all have to be paid for. It is enough to raise a new question: of the historical price, which costs come from physical constraint, and which from the existing way of making things (Musk, 2012).

The starting point of the derivation is basic fact and physical constraint, not a ready organizational standard. Physical relations help a team see what is impossible and where the theoretical floor sits. The organization still has to decide how reliability, safety, delivery, and cost rank, and then verify the derived design in engineering trials. Musk did not "borrow a deeper standard." When no answer that could migrate was on hand, he proposed a candidate from more reliable facts.

6. Once more methods work, why ranking is still required #

The Clarity Method leaves most of the resource for seeking a better solution by means. In the work that means: find several methods around the same aim, keep the first try small, let results knock out the weaker plans, and keep revising on the feedback.

This part takes 80 not because the two 10s do not matter, but because standards and data can only say how to compare. They cannot complete the product, the process, or the service for the organization. Other people can tell you which demands have already appeared, and which risks must be avoided. They cannot combine the resources in front of you into something that can be delivered. Innovation does not have to invent a direction from scratch. It can also take the form of making a concrete method better than the existing choices.

A team may try out three plans that all run: one is fastest, one is cheapest, one best protects user trust. Without a shared scale, the more "feasible" options there are, the easier it is for rank and volume to decide the review.

Chapter 4 already supplied the shared standard of judgment: judge by value. When plans are compared, each candidate has to answer: what observable change it brings, for whom; when a signal is expected; which limiting resource among money, time, and judgment it consumes; and which boundaries of integrity, safety, compliance, and survival cannot be traded.

"Better," then, is not a highest score that floats free of conditions, and still less a promise to find a mathematical global optimum. Under the present aim, evidence, and constraints, it only says that one method is more worth continued investment than the other feasible ones. After the conditions change, the ranking can change too.

Only after verification ends does the team form or revise an organizational standard from the judgment it has: keep the conditions of use, the reason for the choice, the evidence of acceptance, the stop line, and the uncertainties still unresolved. 10/10/80 is also not finished after one pass. After a trial exposes a new angle or a new condition, the team refines the standard again, fills in the key data, and then enters the next round of comparing methods.

7. Paired samples are not two stories #

When a standard is borrowed from outside, successful companies are usually still operating, and the public filings, interviews, and reporting are also more complete. Failed companies may already have left the field, and the public record is usually thinner. From the start, the material on hand leans toward winners.

Jerker Denrell analyzed this systematically in management research: when failure samples are under-collected, some high-risk practices look especially effective among survivors, even when they did not produce better results in the full population (Denrell, 2003). The Clarity Method therefore uses a rule of paired samples: from winners, look for a mechanism by which a practice may work; from losers, look for the conditions under which the same mechanism may fail.

After pooling many small-firm management-training trials, David McKenzie estimated, with a random-effects model, that training raised profits by 10.1% on average, with a 95% confidence interval of 4.1% to 16.1%, and sales by 4.7% on average, with a 95% confidence interval of 0.2% to 9.2%. Many of the single trials were too small to detect an effect of that size, and results differed clearly across programs (McKenzie, 2021). The set supports "outside management knowledge can, on average, bring a limited improvement." It does not support one set of practices for every firm, and still less does it prove that 10/10/80 is a statistically optimal ratio.

The rule is a method for forming and testing a hypothesis. It is not a shortcut that says "find one win and one loss, and cause is established." For the two sides to have comparative value, at least four things have to be done.

1. Define the outcome first. Before looking for cases, write what counts as a win, what counts as a loss, and how long the observation runs. Look at cases first and define the outcome later, and the definition of win and loss will move with the sample chosen.
2. Bound the comparable range. Take cases, as far as possible, from nearby industries, stages, resources, and time windows. A mature consumer brand and a newly founded software firm, even if both succeeded, may not be explaining the same problem.
3. Trace the process. Do not record only what practice was used. Check when it happened, which conditions it depended on, how feedback returned, and whether other explanations could account for the result.
4. Record what is missing. Write from what population the cases were drawn, whether losers disappeared, and which materials cannot be obtained. When the other side cannot be found, the conclusion can still be used in a trial, but it must be marked as a hypothesis still to be tested.

One winner and one loser can propose a mechanism, find a counterexample, and even overturn an absolute claim that "it holds in every case." They cannot estimate how often the practice succeeds in a population, and they cannot rule out other differences between the two firms. Case study is better at answering "how it may have happened, and why it failed here" than at answering, from a handful of stories alone, "how much it raises results on average."

The statistic has to follow the question. When extremes are obvious, the distribution is skewed, and the aim is to describe a typical location, the median is often more robust. When the aim is to estimate an allocated total, an expected result, or data that are approximately symmetric, the mean may be more informative. The U.S. National Institute of Standards and Technology notes that in a skewed distribution the mean, the median, and the mode reflect different aspects, and that no one measure is naturally correct for every question (NIST, current engineering statistics handbook). When each side has only one case, there is no informative "median." Reporting sample size and distribution first, then deciding which statistic to use, is more honest than declaring one kind of number.

8. Meituan Preferred: how far the public record can go #

The public record on Meituan Preferred lists several operating conditions that did not hold as expected, and can be used to test a migration hypothesis. It does not form a strict pair with Duan Yongping.

According to a third-party transcript of a 2024 earnings call, Wang Xing said Meituan Preferred started in early July 2020. Expansion was fast. Meituan's 2020 annual results said that by the fourth quarter Preferred already covered more than 2,000 cities and counties, and more than 90% of cities and counties nationwide. The same passage also said the business was still in early exploration (Meituan, 2021).

The public reason at the time was not only "fight another takeout war." On the 2020 earnings call, Wang described Meituan's development as moving from service retail into goods retail: takeout had supplied a platform, users, and a base of instant fulfillment, and continuing to extend category and geography was listed as an aim. He also called community e-commerce a rare opportunity of many years, and said new e-commerce infrastructure might reach hundreds of millions of users in lower-tier cities and rural areas. The verbatim notes of the call are a third-party transcript. They cannot replace internal project files. They can confirm that the public account connected existing platform capability with a new business opportunity.

Platform traffic, ground operations, supply chain, and fulfillment experience can all migrate, and three conditions still had to be verified again: whether next-day pickup community e-commerce could let revenue per order cover procurement, warehousing, and fulfillment over the long term, and leave a reasonable gross margin; whether, after scale rose, procurement and fulfillment cost would fall as expected; and whether purchase volume and retention after subsidies were reduced would be enough to carry the operation.

The 2023 annual report said growth of the community e-commerce market had flattened, Meituan Preferred's scale had grown more slowly than expected, and it was therefore hard to cut fulfillment cost per order by much; fierce competition also made it hard to raise markup and cut subsidies. The report described the market as tougher than originally expected, and moved later emphasis from fighting for share toward capability, user experience, organic retention, and cutting losses (Meituan, 2024).

Those are exactly the conditions that should have been checked before migration. That the original business had huge scale does not mean the new model can dilute cost in the same way. That the original platform could acquire users quickly does not mean enough demand remains after subsidies stop. That existing logistics can supply a first condition does not mean next-day concentrated delivery, pickup points, and goods retail already have a sustainable gross margin.

After the adjustment, Meituan Preferred still could not keep operating nationwide. Meituan's 2025 interim results confirmed that in June the company exited persistently loss-making regions, and continued to explore next-day pickup plus new forms of community retail only in core regions. Jiemian News later reported that the last remaining region's warehouse was withdrawn on December 15; at publication, Meituan had not responded. The 2025 annual results, released in March 2026, then confirmed in the company's own words that Meituan Preferred had been shut down (Meituan, 2025, 2026; Jiemian News, 2025).

The operating losses Meituan disclosed each year for the new-initiatives segment cannot be booked straight to Meituan Preferred. Added mechanically on rounded figures, 2020 through 2024 come to about 105.2 billion yuan: 10.9 billion, 38.4 billion, 28.4 billion, 20.2 billion, and 7.3 billion yuan (Meituan, 2021–2025, annual results).

First, the segment also includes Meituan Grocery, Kuailv, ride-hailing, bikes and e-bikes, power banks, restaurant management systems, microlending, and other lines. Second, in 2022 Meituan changed the segment definition and restated the 2021 new-business loss from 38.4 billion yuan on the old definition to 35.9 billion yuan. Adding across definitions can only show that the whole new-business combination once consumed a large resource. It cannot put the whole bill on Preferred. Meituan has not published an independent cumulative profit and loss for Preferred (Meituan, 2022, 2023).

The public record can confirm that Meituan extended existing capability into goods retail, and expanded its footprint quickly while the business model was still in exploration; later, scale, fulfillment cost, markup, and subsidy conditions did not hold as expected. The material cannot confirm whether an internal pilot was run, and it cannot confirm whether takeout's way of operating was copied as it stood.

Duan Yongping and Meituan Preferred are therefore not a win-loss experiment that has already identified cause. The Duan sample suggests: borrow first the demand signal the market has already given, and check whether the later firm has the ability to enter later and still make a better product. The Meituan Preferred sample supplies the other side: even when the broad direction fits the strategy, and users do need goods retail, the cost and retention conditions of the concrete model still have to be verified on their own. Together the two sides form a hypothesis that still needs testing: capability and successful experience already on hand cannot prove that the key conditions still hold in a new scene.

9. Experience can propose a standard; it cannot skip verification #

Chapter 3 already showed that experience may, after feedback, form a fast judgment, and that it may also record only that a practice kept appearing, without recording the relation between the practice and the result. Any rule that comes from a person's history can only start as a candidate standard. It cannot become a company-wide standard because the person who proposes it has seniority.

For a piece of experience to become a standard the organization uses in common, it has to take at least four checks. First, can it say what problem it solves, and not only "this is how it has always been done"? Second, can the conditions under which it holds and fails be written down? Third, do outside successes, failures, and counterexamples support that explanation? Fourth, if another person uses it under nearby conditions, can a result that can be checked be obtained?

Experience can help a team see an anomaly earlier, and it can also propose a hypothesis that outside material has not yet recorded. Experience can propose the next candidate standard. Whether it becomes an organizational standard is decided by the result of verification.

The check matters especially when a judgment is about to be given to a team or to AI for repeated use. That one person, on tacit experience, occasionally patches a vague requirement does not mean the organization already has a standard. Only when conditions of use, the basis of the choice, the falsifying signal, and the result of verification are recorded does personal experience begin to become a shared standard other people can reuse and revise.

10. Reflection finds the gap; summary checks whether it is clear #

A person using the Clarity Method needs two habits. Reflection finds gaps in what has already happened: what this choice rested on, which hypothesis the result overturned, and which angle was still not looked at. If reflection has not produced a candidate standard, a revision, a hypothesis still to be tested, or a next action, it has not yet entered the standards cycle.

Summary states the judgment that was reached. A person has to be able to say what was borrowed, which conditions were the same, which conditions have already changed, and what result would force a change of answer. Summary is not a way of compressing the process into one elegant conclusion. It is a way of letting a second person check the judgment and keep using it.

The Feynman technique only checks whether the summary is clear. State a judgment in ordinary language to someone who does not know the background, and ask that person to restate the conditions of use and the stop signal. If the other person can only repeat a slogan, the key relation is still not clear. Return to the material and the trial, fill the gap, and summarize again. Concrete scheduling and daily methods of reflection wait for Chapter 11.

The three kinds of evidence answer different questions. Duan Yongping and Meituan Preferred show a possible mechanism and conditions of failure; they are not a paired sample from which an overall effect can be computed. The Mexican firm trial and the management-training meta-analysis supply a quantitative contrast; what they test is consulting or training, not the Clarity Method as a whole. Together they support a narrower judgment: after diagnosis, migration, and implementation support, outside management knowledge can improve results for some firms, and the effect differs by firm and by practice. 10/10/80 remains resource discipline waiting for practice to keep revising it. It is not a ratio already shown to be generally optimal.

The question Chapter 6 left—"where does a standard come from?"—can be answered this way. Find which kind of judgment you are missing; look first for a candidate standard that has already been tested against reality, and bring back the angle, the scale, and the conditions; when migration does not hold, turn to bounded exploration; after several feasible methods have been tried, rank them by value and form or revise an organizational standard. What that expands is the organization's own options.

More options are still not enough. How other people see and understand your plan, and on what basis they choose among several plans, is another class of problem. Chapter 8's Interference Method continues with that.

What to Do Monday Morning #

Pick one piece of work that is about to be expanded, and that the team is still arguing over. Do three steps first.

1. Write the gap. Define what counts as a win and what counts as a loss, then judge whether what is missing now is the angle of observation, the scale of judgment, or the conditions of use. Do not finish the cases first and then change the definition of the outcome.
2. Form a hypothesis from paired samples. From nearby conditions, find one success and one failure, and record a possible mechanism, the conditions of failure, and the source of each. If one side cannot be found, or the two sides are clearly not comparable, mark the present conclusion as a hypothesis still to be tested.
3. Run a migration check and a small trial. Set the present object, stage, resources, constraints, and way of feedback beside the original. Write the range of this trial, the evidence of a pass, and the stop condition. If several feasible methods appear, rank them by "what value, for whom, when it pays off, which resource it consumes, and which boundary cannot be crossed."

Finally, have someone who does not know the background restate: why this method is being tried, under what conditions it holds, and what result would require a stop. Where it cannot be stated, do not cover the gap with more terms. Put it on the list of data and verification the next round still has to fill.

Chapter Acceptance Self-Check (against chapter acceptance standards) #

1. Claim restatable in one sentence ✓: 10/10/80 is resource discipline, not a statistical law; in most cases borrow a verified candidate standard together with angle, scale, and conditions of use, and test it against reality.
2. Ten-section spine and whiteboard figure ✓: Chapter 6 handoff → 10/10/80 resource discipline → dimension / standard / conditions → borrow and migrate → Duan → Musk as exploration branch → rank by value → paired samples → Meituan Preferred → experience and personal method. Figure is resource bar plus action chain plus evidence bound; 40/40/20 is out.
3. Evidence caliber ✓: Duan from 2018 Stanford public compilation and 2025 Zhejiang University transcript; BBK VCD from CCTV 1997 ad and quality rating, not a 1996 entry. Musk from Henry Ford 2008, TED 2013, WIRED 2012 ~2% as his estimate, not a proof that a product should sell at materials cost. Meituan Preferred from company filings on coverage, cost, markup, subsidy, June 2025 exit, and 2025 shutdown; December 15 from Jiemian News with no company response then; ~105.2 billion yuan as a cross-definition mechanical sum of the new-initiatives segment, not booked to Preferred. Bruhn, Karlan & Schoar 2018 and McKenzie 2021 as adjacent quantitative support, not a test of 10/10/80.
4. Case contract ✓: Duan for method and conditions of use, not operating cause; Meituan Preferred for conditions that did not hold as expected, not a copied takeout script; Musk as exploration when nothing can be borrowed, not "borrow deeper"; author logo, 40/40/20, and "most answers already exist" are out.
5. Paired samples and statistics ✓: Denrell 2003 on missing failure samples; one win and one loss can propose or revise a hypothesis, not prove cause; mean and median follow the question and the distribution, per NIST; no median iron law.
6. Chapter boundary ✓: empiricism compressed to organizational checks; Jobs, Johnson, and the author logo out of the main argument; reflection and summary as method only; calendar left to Chapter 11.
7. Fluency ✓: rewritten in English voice from the Chinese authority; current prose-standard.