Chapter 5Limiting Resources: Money, Time, and Judgment

1. Two pieces of paper: the budget is closer to the real strategy than the declaration #

A company's annual strategy is usually easy to find. It may speak of growth, innovation, and customers first. It may put AI in the most visible place. Another document is not always public, and it sits closer to the actual trade-off: who received budget, which roles were allowed to grow, what the leadership kept whole blocks of time for, and where cash finally left the accounts.

The two documents do not always confirm each other. A stated priority that has no owner, no trial budget, and no management time has not yet entered resource allocation. A line of work with no loud name, if it keeps getting the main budget year after year, already is the real strategy. Where resources flow cannot prove that every dollar was spent well. It can show which choices the organization made by giving up others.

Budgets also distort. A department may grab quota first. A project may receive money and never start. Accounting categories can hide what an outlay was actually for. Budget, actual spend, headcount, and the principal leader's time have to be read against one another. Only when several signals keep pointing to the same place can you name the company's actual order of priority.

At the company layer, money matters for more than what it can buy. Cash also decides how many failed tries the organization can bear, how long it can wait, and whether it can still act after one miss. Short-term urgent spend keeps today alive: wages, collections, delivery, outages, and debts coming due. Leave them untreated and the organization bleeds at once. Long-term important spend builds tomorrow: new capability, customer trust, judgment quality, and reusable work. Doing a little less of that today may not set off an alarm. A few years later, the gap shows.

Both kinds of spend can be right. When cash is about to run out, staying alive is not short-termism. When cash is ample, and every resource still goes to putting out fires, long-term building never gets funded. Every finite resource needs a written account: why it goes here, what it is meant to buy back, and under what conditions the organization will stop or turn.

Judgment can become an organizational asset. It does not increase on its own. Money, people's time, and reliable judgment are all limited. Before you allocate them, you have to see where the bind is now.

2. Three levels of analysis: the limiting resource migrates #

In a chemical reaction, the reactant that is used up first sets how much product can form. Piling on more of the other ingredients will not raise that ceiling. A limiting resource is the resource that, under present conditions, most tightly constrains the output you are after.

"Present" cannot be dropped. Before financing lands, a company may be unable to keep running trials because cash is short. After the money arrives, the bottleneck may move to customer demand, a regulatory permit, a key hire, or the quality of judgment. When a product enters scale, a technical capability that used to be scarce can become ample supply, and delivery and channels start to limit growth. A limiting resource names a constraint in one stage. It is not a permanent identity for the company.

Company, person, and AI-native organization are three levels of analysis. Money, time, and judgment are not mutually exclusive answers to the same question.

At the company, look first at money. Markets, talent, and technology can all become bottlenecks in a given line of work, and they still land on the same resource table: how much the organization still plans to put in, how long it can keep putting it in, and whether a next attempt remains after a miss. Money is the most general budget constraint at the company layer. It does not follow that every problem can be solved by spending more.

At the person, look first at time. Money can be earned again. A task can be handed off. The hour that has already passed does not return to the account. A person can trade time for money, for learning, or for relationships, or keep it for thinking. However the trade is made, what is paid is the same non-renewable resource. Choosing one thing means giving up the other things that could have occupied the same stretch of time.

At an AI-native organization, look first at judgment. What is scarce is not how often people declare a stance in a meeting, and not only a few executives' hours. It is the capacity to produce reliable judgment: facing an important trade-off, being able to say what value ranks first, which facts the call rests on, and what result is expected, and allowing the actual result to overturn the original answer. Doing more judging does not, by itself, raise capacity. Ten unverified opinions may still be worth less than one judgment that can be reviewed.

The three layers join into a loop. The company's money buys people's time. People's time carries judgment. Judgment then decides where the next round of money and time will go. A wrong judgment takes the first two layers with it. When cash runs out, even a good judgment loses its chance to act. When a person's time is filled with chores, judgment has no room in which to be produced.

Figure: money, time, and reliable judgment are three levels of analysis, not irreversible historical stages. The current bottleneck migrates around the loop.

That loop also explains three common mismatches. Money can buy information, tools, and talent; it cannot complete the trade-off for the organization. Longer hours can raise the number of tries; they cannot make a wrong direction right on their own. If reliable judgment that has already been formed is spent on sorting, retrieval, and formatting that can be automated, the scarcest capability is being spent where scarcity is not. Money, time, and judgment can support one another. They cannot impersonate one another.

3. Clear the baseline first, then ask about ROI #

When a resource is currently limiting output, putting it into A means it cannot, for now, go to B. Its cost is not only what appears on the payment voucher. It also includes the option given up. Asking for a return on a limiting resource is not a way to turn everything into profit. It is a way to put the cost of the choice into the calculation.

ROI, though, cannot sit in front of every rule. Legality, integrity, and safety are boundaries that a high return cannot offset. Cash and solvency decide whether the organization is even eligible for a next round of action. A plan that is expected to earn more still should not enter the comparison if it depends on deceiving customers, breaking regulation, or taking a risk the organization cannot survive. First exclude what cannot be done. Then discuss return among the plans that remain.

"Org-wide people-investment ROI > 1" uses a multiple of value to cost, which is not the formula most finance textbooks lead with. Ordinary financial ROI is net gain divided by cost: you subtract cost from the return first, and break-even is zero. The formula used here is:

Limiting-resource ROI = verifiable total value in the agreed window ÷ full cost in the same period.

On this definition, a result above 1 means value covered cost; below 1 means the outlay has not yet been recovered. For the same set of values and costs, ordinary financial ROI equals this multiple minus one. The two definitions cannot be mixed, and the multiple is not written as a percentage.

The hard part is not the division. It is when the calculation happens. Before the outlay, the organization writes a hypothesis: what value it means to create, for whom; how much resource it expects to consume; and what signal it expects to see in which window. After the outlay, it checks value, cost, timing, and attribution against what actually happened. The version written in advance and the result afterward have to be kept side by side. If the definitions change once the results are in, the review no longer does any work.

Long-term value is under the same discipline. Revenue, gross profit, and cost savings that can be confirmed can be measured directly. Brand, trust, capability, and standards cannot be turned into money at will just to fatten a numerator. The more honest move is to write, before the outlay, observable proxy signals, a standard of evidence, and a date to look again. A proxy cannot prove that the final value has already arrived. It can let reality tell the organization, mid-course, whether the original path is still worth paying for.

So ROI is not short-termism. Counting only cash in front of you will miss future capability. A long project with no mid-course signal, no review date, and no stop condition can consume resource without a limit. The later the payoff, the earlier the acceptance standard should be written down.

Amazon's 1997 shareholder letter is a rare physical object: a young company committing in public both to long-term investment and to acceptance and exit.

4. Amazon: a long-term ranking still needs acceptance #

Amazon attached its first letter to shareholders after going public to the 1997 annual report. One of the headings inside it is It's All About the Long Term. That year, sales were $147.8 million, and headcount rose from 158 to 614. The company was growing quickly, and it was still losing money on a net basis; year-end cash and investments still included proceeds from the offering and from loans. In other words, this was not a company so flush that it could ignore cost. It was a young company that had to tell shareholders how the money would be used.

The letter first set a ranking. Amazon would make investment decisions in light of long-term market leadership, not short-term profit or Wall Street's immediate reaction. When it had to choose between optimizing GAAP accounting and maximizing the present value of future cash flows, it chose the latter (Amazon, 1997). The object of evaluation moved from the current income statement toward future cash flow and market position.

The same list also wrote acceptance. The company promised to measure each project and investment by analytic results, to abandon those that could not provide an acceptable return, and to put more into those that worked best. Where there was a sufficient chance of a market-leadership advantage, it would take more risk. Some investments would succeed and some would not, and both results had to produce learning (Amazon, 1997).

The two sets of clauses together govern capital allocation. The long-term ranking names what the organization is willing to wait for. Analytic acceptance names the conditions under which it will not wait forever. Allowing a single project to fail is not the same as allowing a project to have no falsifying condition. Declining to chase quarterly profit is not the same as declining to ask about return. Long-termism and ROI answer different questions: the span of time, and the acceptance of value.

The following year, Amazon again attached the 1997 letter, and raised the print run from about 13,000 copies to more than 200,000. Bezos wrote that the company did not claim this was the correct philosophy, only that it was theirs. The material can confirm the print run and a ranking that stayed public. It cannot confirm an exact shareholder count, and it cannot prove that every investment was executed strictly by the letter (Amazon, 1998).

That public restatement continued for many years. The 2020 shareholder letter still carried the 1997 original, and looked back at Prime, Marketplace, Alexa, and AWS. None of those businesses was even an idea in 1997, and none was destined to succeed; the company had taken risk on each. By the end of 2020, AWS was running at a $50 billion annualized revenue run rate (Amazon, 2020).

Those later facts show that a long, high-risk investment can produce value far outside the original business boundary. They cannot be run backward to prove that the 1997 letter "caused" AWS, and they cannot be forced into a growth multiple that pairs AWS's annualized run rate with the whole company's 1997 annual sales. A shareholder letter is management's public commitment, not a causal experiment. Successful businesses also move with technology, markets, talent, and luck.

The material supports a narrower conclusion. Long-term capital allocation can have a stable ranking, and it can at the same time specify what return is acceptable, when to stop, and what a failure has to leave behind. It cannot prove that one shareholder letter is enough to produce the same operating result.

5. Evergrande: the public record does not present one shared acceptance standard #

Evergrande Mineral Water Group was formed in 2013, and Evergrande Ice Spring entered the market that November. At a national partners' conference in January 2014, Xu Jiayin set a sales target of 100 billion yuan for the year. The next year's results meeting disclosed 2014 sales of 1.09 billion yuan and a loss of 2.37 billion yuan, and explained the loss as upfront spend on advertising, national distribution, and building a sales force (China Evergrande, 2015 results meeting).

Those two figures are not independently audited Ice Spring segment numbers. China Evergrande's listed-company annual report folds mineral water, hotels, football, property construction, and other lines into an "other businesses" segment. The 1.09 billion and 2.37 billion yuan come from management disclosure at a results meeting. Income and loss figures in later filings by related operating entities also differ because the entity scope differs. They can show a large gap between the target and the company's self-reported result. They cannot stand as an independently audited conclusion on a single definition.

Evergrande did later exit. In September 2016, China Evergrande announced the sale of all interests in grain and oil, dairy, and mineral water for a total of 2.7 billion yuan, giving as its reason a wish to concentrate more on real estate and related businesses (China Evergrande, 2016). The exit shows that the company would cut a loss. The announcement does not explain three things: what value hypothesis those businesses were opened on, what shared signal was used to review them along the way, and how the next round of investment judgments was revised after the exit.

Resources then kept moving in other directions. Haihua Island is often written up as "160 billion yuan of cumulative investment," but 160 billion yuan is the planned total investment the company published. In a 2022 open letter, Evergrande Children's World Group said actual cumulative investment over more than six years of construction was about 81 billion yuan (Evergrande Children's World Group, 2022). Evergrande Auto's management, at the 2021 results briefing, disclosed cumulative investment of 47.4 billion yuan in the new-energy vehicle business to that date, some of it for technology acquisitions and R&D, some for plants, equipment, and parts (Evergrande Auto, 2021). Both figures are company self-reports and should be used as such.

One item can be read straight from the annual report. Evergrande bought Vanke shares in 2016 and sold the entire holding to Shenzhen Metro in 2017, finally recognizing a disposal loss of about 7.176 billion yuan. That is a realized loss, not a "paper loss" still moving with the market (China Evergrande, 2017).

The later ledger shows what the company absorbed. As of the end of 2022, China Evergrande disclosed total liabilities of 2,437.41 billion yuan, including 721.02 billion yuan of contract liabilities. Net losses were 686.22 billion yuan in 2021 and 125.814 billion yuan in 2022, about 812.03 billion yuan together (China Evergrande, 2023). On January 29, 2024, the Hong Kong High Court held that the company had not put forward a viable restructuring plan supported by the required majority of creditors, and issued a winding-up order (Hong Kong High Court, 2024).

Those numbers cannot be strung into a single-cause chain in which the lack of a unified ROI made Evergrande fail. High leverage, a changing property market, regulation, governance, the financing environment, and project execution all play in the result. The absence, in the public record, of one ROI acceptance used across mineral water, cultural tourism, automobiles, and equity investment does not prove that no such system ever existed inside. The material on hand is not enough to conclude whether an internal system was present.

When resources cross businesses that differ widely, a public, traceable value hypothesis, review window, and stop condition can be used to tell whether a later outlay absorbed the last failure. Evergrande's public record is not enough to make that distinction.

Amazon and Evergrande are not a same-industry comparison, and they are not an experiment that held other variables fixed. They can be compared on one item of capital allocation. One side made public a long-term ranking, an acceptable return, and a promise to exit. The other side's public record contains a large volume of targets, outlays, and results, and is still not enough to reconstruct an acceptance ruler used in common across businesses. The difference supplies a mechanism. It does not supply a final assignment of blame.

Chapter 2 already recorded that Klarna, after overweighting cost, saw service quality fall and therefore put high-quality human support back in. Counting only cost saved, and leaving customer value out, produces an incomplete return.

People investment is harder to attribute than a single project. Output is often produced by several people together, and the value may appear late. Building a precise profit-and-loss statement for each person disguises attribution that cannot be confirmed as if it were a number.

6. Org-wide people-investment ROI > 1: an investment frame, not a personal ranking #

Reliable judgment is usually made first by a person, so an organization that wants more judgment often has to hire, train, and keep people. It still does not follow that every person should have a precise ROI. A product's success may come at the same time from research, design, engineering, sales, and support. The value of finance, legal, security, and infrastructure shows up through other people's work. Forcing a shared result into personal numbers only hands the power of allocation back to subjective judgment.

"Org-wide people-investment ROI > 1" constrains the organization's duty to invest: every investment in people needs a value hypothesis written in advance, and a review at the agreed time. The unit of acceptance does not have to be the individual. Where a result can be attributed directly, the person can be the unit. Where several people deliver together, the team or the project should be. Shared functions such as finance, legal, platform, and management should be reviewed as a service portfolio and the business results they support.

A usable investment card has to answer at least five questions.

First, who is the object of value. Not a vague "raise efficiency," but which customers, which team, or which critical result will change, and how. If the object is unclear, the metrics that follow are only measuring activity.

Second, what full cost includes. Pay and benefits are the start. Recruiting, onboarding, tools, direct management, and coordination still have to be listed. Opportunity cost is stated separately: which feasible option the organization gave up for this outlay. It cannot be counted as cost and then subtracted again as "revenue not earned."

Third, at which unit acceptance happens. Salespeople can sometimes be reviewed as individuals. A product feature usually has to be reviewed as a cross-functional team. Security and legal are better reviewed by risk control and service results. Choose a reasonable unit first, then talk about attribution. Do not decide in advance that every person must be scored, and then force a team result down the roster.

Fourth, how attribution is done. Direct closed business, clearly identified savings, and independent trials can support stronger attribution. Joint projects, brand, and management improvement usually support only partial attribution. The organization should separate direct, shared, and supporting contributions, write down the hypothesis and the uncertainty, and not use decimal places to hide what it does not know.

People investment can in fact cross a person's boundary. A randomized field experiment in a Dutch call center found that employees who took part in training raised productivity by about 10%. For every 10 percentage-point rise in the share of trained colleagues on a team, untrained colleagues' productivity rose by about 0.51%, and service quality did not fall (De Grip & Sauermann, 2012). The result holds under those experimental conditions. It does not prove that all training has the same return. It does make the attribution problem concrete: counting only the people trained will miss team gains; booking all team gains to the people trained will overstate personal contribution.

ISO 30414:2025 places human-capital ROI among productivity indicators at the organizational layer, rather than building a causal score for each employee. Its formula is not the value-to-cost multiple above, and it cannot prove "org-wide ROI > 1." What it does show is that choosing organization, team, or person as the unit of acceptance is a question that has to be answered before the arithmetic.

Fifth, when to continue, adjust, or stop. Before the outlay, write leading signals, a review date, and a stop condition. Stopping is not only cutting a person. It can also mean changing the goal, filling in context, shrinking the trial, regrouping the team, or cancelling the project. The ROI card's job is to improve resource allocation. It is not to find a number that will justify a people decision already made.

Full cost is especially easy to replace with a flattering multiple. The U.S. Small Business Administration's rule of thumb is about 1.25 to 1.4 times base pay. In Bureau of Labor Statistics data for March 2025, wages were about 70.3% of private-sector employer compensation, or about 1.42 times when wages and benefits alone are converted (SBA, current page; BLS, 2025). Those figures belong to particular countries and statistical definitions. They also leave out office, management, and opportunity costs. They cannot be turned into a universal multiple for a Chinese company, and still less can they prove that a fixed 1.7× or 2× is true full cost.

Calculating from the actual books is less convenient than a fixed multiple, and more reliable. Costs that can be confirmed directly are listed item by item. Costs that have to be allocated come with a written rule. Costs that cannot be calculated reliably remain as uncertainty. What a manager needs is an estimate good enough to support a choice, not a table that looks precise and cannot be audited.

Value has to be layered too. Confirmable revenue, gross profit, and savings can enter the multiple. Long-term capability, customer trust, and reusable judgment first enter an evidence column, with proxy signals and a path to realization written down. When the evidence is enough and financial results appear, the actual ROI is updated. Leaving a contribution that cannot yet be monetized in the evidence column keeps the numerator from being inflated by an invented price.

Regulatory, safety, and integrity roles make the same boundary plain. Their work sometimes consists in "the bad thing did not happen." That is hard to assign precisely to one person, and the absence of an incident in the present period is not proof that the outlay was surplus. The organization should still accept response time, audit results, material defects, recurrence, and the quality of the business supported. It should not, in order to satisfy the same formula, put an arbitrary price on a disaster that did not occur.

So org-wide ROI is an investment frame, not a ranking. Founders, managers, and shared functions all take the same value review. Where attribution cannot be confirmed, uncertainty is kept. A long investment in people cannot be renewed without limit in the absence of review evidence.

The standard also needs a time dimension. The same outlay can look entirely different at three months and at one year. Without a window, long-term value is killed too early, or postponed without end.

7. 3, 6, and 9–12 months: a long project still needs a review date #

The default validation windows come in three bands: early, 3 months; mid, 6 months; later, 9–12 months. They are not industry averages from a statistical study, and they are not a probation period every role must obey. They are an internal management rule. The earlier the organization, the faster cash and direction change, and the sooner it needs evidence enough to continue or stop. Once the organization is further along, product, brand, regulation, and cross-team work take longer cycles. Long results can be given more time. Mid-course review cannot be cancelled.

Early, mid, and later name the organization's stage, not a person's tenure. An early organization's business model and sources of cash are still being tested, so the default is a formal review at three months. A mid-stage organization already has a more stable business base and is still building new capability, so the default is six months. A later organization's outlays often run through a budget year, a complex system, and a regulatory cycle, so nine to twelve months can be used. If a given project has a shorter natural cycle, shorten the window. Drug development, large hardware, or a regulated project with a longer cycle should be adjusted to real milestones.

A validation window is the longest the organization will wait before a formal reckoning. It is not a protected period in which the outlay is guaranteed to continue. Every stage should watch leading signals by the month. If the signals keep diverging and no new credible evidence appears, review early. If the signals look good for now, that only means the path has not yet been falsified. It does not mean the final value has already arrived.

A leading signal has to explain how it leads to the final value. A sales outlay can watch whether qualified opportunities move into later stages, not only how many calls were made. Product infrastructure can watch incidents, delivery speed, and reuse, not only how much code was committed. A brand outlay can watch whether target customers form repeatable, unprompted search and referral, not only exposure. Activity is easy to grow. A value signal is harder to fake.

Long projects especially need the final result and the mid-course evidence kept apart. The final result may not appear for a year. Mid-course, the organization should still know whether the most important hypotheses have held, whether cost has overrun the original estimate, and whether outside conditions have already changed. Review does not demand that the project deliver its full value early. It decides whether the organization still has reason to keep paying.

In a 2016 Caijing interview, Zhang Yiming said he would keep enough cash in reserve and push some spending "a bit further back," in order to raise the standard and leave a buffer (Zhang Yiming, 2016).

In the same interview he also said that if Toutiao had spent a little more in 2012 and 2013, growth might have been faster. He split missed timing into two kinds: delayed-gratification allocation done badly, and a failure of judgment. The self-report records both the return to waiting and the opportunity cost of waiting.

The material is an interview self-report, not audited causation. It supports recording, in a cross-period choice, both future return and the opportunity cost of the wait. It does not support a specific personal time mix.

After the window closes, the review produces more than "continue" or "stop." It should also leave a new understanding: where the original hypothesis held, where it failed, which signals arrived earlier than expected, and which costs were missed at the start. Those results, once they have met reality, are what make people investment different from a one-time consumption of machines and cash.

8. Why people get stronger: verification, reflection, settling #

Finishing a piece of work does not, by itself, make a person stronger than before. Repeating the same method for ten years may only be replaying the first year's experience nine times. Accumulation takes three moves: the result is verified, the person who judged looks back, and what the looking-back produced is written down.

Verification puts the judgment written in advance next to the actual result. If the result matches the expectation, it still has to confirm which conditions did the work. If it does not, it has to separate what happened to the original goal, to the key conditions, and to the execution of the action. Without that step, personal capability, execution, and the environment cannot be told apart.

Reflection has to change the next action. It turns the result into knowledge that can be used again: which signals should have been checked earlier, which boundary cannot be crossed, under what conditions the original method holds, and which weakest hypothesis should be tested first next time. A postmortem that does not form a revised judgment or action cannot enter the next round of verification.

Structured reflection already has external evidence on later performance. Tannenbaum and Cerasoli pooled 46 individual and team debrief samples, 2,136 participants in all. Against controls, a correctly run debrief raised later performance by about 20%–25% on average, with similar effects in individuals and teams, and in simulated and live settings (Tannenbaum & Cerasoli, 2013). The meta-analysis supports debriefs that have a goal, a discussion, and a change in action. It does not support treating an unstructured look-back as equally effective, and it cannot supply a statistical basis for the three validation windows.

Settling writes the verified result into a carrier that can be checked. At first it may be only a judgment record, a check condition, or a hypothesis that has been falsified. The next time a similar problem appears, the record should say why the last choice was made, and what result once overturned it. A record that other people and AI can keep checking, using, and revising is what constitutes the long-term value of people investment.

After those three moves, people investment gains an accumulating advantage that one-time resources do not have. Money, once spent, has to be earned again. Time, once passed, cannot be retrieved. A reliable judgment, once verified and left with a result, can raise the starting point of the next choice. A judgment that has been screened by reality can shorten the handling of a similar task, and it can support a more complex one.

That advantage also explains why people ROI cannot look only at output in the present period. If a delivery also leaves verified conditions of judgment, and reduces the team's repeated trial-and-error later, the value is more than one present result. The later part has to be actually recorded, reused, and verified. A vague "growth" cannot be booked.

Accumulation does not guarantee that capability will rise forever. Environments change, and old methods go stale. Only by keeping new results coming in, and revising old judgments, can the advantage keep widening. Years of tenure record how long time has passed. Verified judgments record how much capability has increased.

When a person's judgment leaves no record, the organization still has to pay the cost of the same kind of judgment again. After the original judge leaves, the unrecorded conditions, evidence, and boundaries leave too. Only when the judgment is written as a standard that other people and AI can check, use, and revise can personal accumulation become organizational accumulation.

What to Do Monday Morning (principal-leader view) #

Start with the ten largest resource commitments of the last ninety days. They can be hiring, procurement, product projects, and market spend, or matters that have been occupying management time at the top. Rank them separately by cash outlay and by key people's time, then merge and take the top ten. Do not sort by department first. First write down the money and time each item actually consumed.

Then answer eight questions for each:

1. Whose situation is this outlay meant to improve, and how?
2. What is currently limiting the result: money, time, reliable judgment, or another concrete step?
3. What are the legal, integrity, safety, and survival boundaries?
4. Which full costs can be confirmed directly, and how will allocation and opportunity cost be handled?
5. Should acceptance sit at the person, the team, the project, or a shared function?
6. What is the final value, can it cover full cost on the present estimate, and which leading signals will be watched mid-course?
7. Given the organization's stage, does the formal review sit at 3, 6, or 9–12 months, and why does the project's cycle require an adjustment?
8. What evidence, if it appears, means continue, revise, or stop?

Pick any one item and freeze the answers as the version written in advance. Each month, update only actual evidence; do not go back and change the original definitions. At the formal review, put expectation and actual side by side, and record the error in judgment. Where a shared function has no reliable personal attribution, accept at the team or service-portfolio layer. Do not invent personal numbers in order to fill a form.

Finally, set the company's publicly stated priorities next to these ten commitments. If the company said AI first, money, people's time, experiments, and management attention should show a change commensurate with that ranking. If none of the four has moved, the strategic declaration and actual resource allocation are not the same thing.

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

1. Claim restatable in one sentence ✓: look first at money, time, or the capacity to produce reliable judgment, depending on the level; the binding constraint migrates; hold the baseline, then accept with a value hypothesis written in advance.
2. Eight-section spine and whiteboard figure ✓: two papers → three levels and migration → ROI after the baseline → Amazon → Evergrande → org-wide people investment → validation windows → accumulation and the handoff to Chapter 6. Figure is an allocation loop, not an irreversible funnel.
3. Evidence caliber ✓: Amazon 1997 ranking plus acceptance, 1998 print run not shareholder count, 2020 AWS run rate not a 1997 growth multiple. Evergrande Ice Spring as management disclosure; Haihua Island planned versus self-reported actual; Vanke as a realized loss; no single-cause chain. De Grip & Sauermann; Tannenbaum & Cerasoli; ISO 30414:2025 as a unit-of-acceptance boundary, not a proof of this book's multiple.
4. Case contract ✓: Amazon for ranking and acceptance stated together; Evergrande for a public record that cannot reconstruct a shared ruler; Zhang Yiming and Klarna as echoes; MD Anderson out.
5. Formula ✓: value-to-cost multiple, not net financial ROI; ex ante and ex post kept apart; shared functions accepted at team / project / portfolio; SBA and BLS figures not turned into a China multiplier.
6. Chapter boundary ✓: no Clarity Method, Interference Method, standards cycle, or personal calendar in the body; individual ranking removed.
7. Fluency ✓: rewritten in English voice from the Chinese authority; current prose-standard.