Chapter 12Reverse Iteration from Consensus: Evolving the Product Path

1. After people and time are in order, where the judgment goes #

Chapter 11 rearranged the calendar. For a relatively stable, judgment-primary job, more of the day should go to judgment, writing, developing people, and learning, rather than being filled entirely by the matter in front of you. Opening six hours on the calendar does not, by itself, guarantee that the organization will then get the product right. The new judgment time can also enter a direction that still has no evidence of demand, or add a technical display that does not add a customer result.

A product path therefore has to answer three things: which class of demand to begin from, what evidence decides a continuation, and how much still-unverified uncertainty the present capital and capability can bear.

Duan Yongping's starting point is already in Chapter 7: let the market prove the demand first; a later firm still needs the ability to make the product well. That public remark cannot prove that later firms succeed more easily as a class.

Meta began from another starting point. Horizon Worlds asked people to put on a headset and, as avatars, enter user-created worlds to socialize, play, and do things together. The company had not only to ship a product. It had to prove that enough users would repeat that experience. By 2022, retention, monthly actives, and internal feedback had not completed that proof. Meta could subsidize the exploration for a long time with profits from other businesses. Horizon Worlds therefore did not hit a cash stop line as quickly as an ordinary startup product would.

The pair does not support "caution always wins, innovation always loses." BBK entered a mature category and still had to face strong incumbents. Meta chose a frontier direction, and that does not prove that a frontier direction is bound to fail. What can be compared are three observable facts: what evidence each company began from, whether the capability already on hand could support the path chosen, and what results decided the next step.

Treating originality as value, and a technical display as progress, bypasses demand, delivery, and survival. Originality can create value. "Nobody has done this before" is not a reason a customer pays. Technology can run ahead. A lead in technology cannot stand in for operating evidence.

2. Choosing a line of work, answer two questions first #

Before you choose a forward path or a reverse path, first judge whether the matter itself is worth doing. The judgment does not rest on a formula. It rests on two questions that both have to be answered.

The first question is: why now?

The answer comes from a variable. A rise in model capability, a fall in the price of inference, a change in a regulation, a reallocation of channels, or a maturing supply chain can all open a window on something that could not be done, or could not be done at a profit, before. A variable says what changed in the conditions. It does not say why a user will necessarily buy. AI can move part of a service's execution from human work to compute. It cannot, by itself, manufacture a use and a payment that did not exist.

The second question is: why will it still be worth doing in ten years?

The answer comes from an invariant. Chapter 6 already borrowed Bezos's judgment: long-term investment should rest on relatively stable customer demand. Amazon's 2008 shareholder letter wrote low prices, vast selection, and fast, convenient delivery as demands that would remain stable for a long time, and used that to explain continued investment (Amazon, 2009). What can migrate is not those three answers. It is the form of the question: what result will make customers choose and pay, again and again, over a long period?

Both questions have to be answered together. They cannot be multiplied. Variable and invariant have no common unit, and there is no verifiable number that can sit on both sides of an equals sign. If one of them cannot yet be written clearly, the business is still sitting at the stage of a hypothesis. That cannot be written as a mathematical law that "if either term is zero, the opportunity is zero."

AI and hardware can show how the two questions come apart. Changes in models, sensors, and the cost of inference may open a new window of interaction. That is a variable. People in a car, at the ear, at a desk, or on a factory floor still need an easier way to get information and finish an action. That is a candidate invariant. Whether a concrete product holds still has to be checked, item by item, for burden of wear, battery life, supply chain, channel, price, and frequency of use. The split does not support a conclusion on any AI-hardware track or any investment.

When bioby.ai chose an overseas influencer-marketing Agent, it used the same structure of judgment. On the variable side, finding people, screening, outreach, follow-up, and record-keeping can be reorganized by models and workflows. On the invariant side, a merchant entering a new market still has to obtain trusted attention. Those two sentences can only explain why it was worth a try at the time. They cannot prove that the product has already run. Whether anyone pays on a continuing basis, whether delivery is stable, and whether unit economics hold, all have to be answered by later operating results.

Only after the two conditions have been screened does the product enter the choice of path.

3. Locate the demand first, then locate yourself #

The shared principle of Part IV is dynamic lifecycle management: locate the object, locate yourself, then match and choose. On a product, the object is the target demand. You are the present organization.

Locating demand cannot let spoken preference in an interview stand in for behavior. More reliable clues include: does this problem keep recurring? Are users already paying for an old solution, or continuously spending labor, time, and risk to work around it? If the technical introduction is taken off, does the problem and its loss still sit there? The more stable those pieces of evidence, the more mature the problem consensus. What is mature is the problem and the expected result. It does not mean that the present solution is already good enough.

Locating yourself puts four ledgers on the same page. Capital decides how long exploration with no revenue can be borne. Channel decides whether the target user can be reached at a reasonable cost. Technology decides what can be delivered reliably now, not the best result once seen in a lab. Organizational capability decides whether sales, implementation, support, and feedback can become a repeatable action. Miss any one of them and a product direction is mistaken for a purely technical choice.

Only after that locating is done is a path matched.

A forward path starts from a demand that is not yet mature, or from a new use of a technology. It fits conditions of this kind: the technology window is short, and a standard, a key supply, a license, or a strong network effect may lock in soon; the team does have a lead on a key technology; early customers have already put time, budget, or data into the verification; and there is enough capital, and an independent source of survival, to bear category education, a long R&D cycle, and several rounds of failure. The premise of a forward path is that the team can pay for uncertainty.

Consensus reverse iteration starts from a mature problem. It fits another set of conditions: users have already proved, by real behavior, that the problem exists; a new technology can improve the result through a familiar experience with a low cost of learning; the organization needs revenue and feedback earlier; the channel, service, and support cost of mature users can still be borne; and the technology window is not so short that a later firm loses the right to choose at the start.

The technology-adoption life cycle uses another scale. Rogers's five adopter categories classify people by the relative time of adoption of a given innovation in a given social system. A continuous variable of time is cut into five ideal types: innovators, early adopters, early majority, late majority, and laggards. It is not a person's permanent character, and still less a rank of capability, value, or product (Rogers, 2003). The same person can adopt one tool very early and another tool very late.

The "chasm" Moore named is the market transition between early adopters and a pragmatic early majority. The former can accept an incomplete product for a vision. The latter require a reliable complete product and proof from similar customers. It does not mean that every added capability has to cross a chasm, and it cannot be used to explain college-entrance bands, pay grades, or personal growth (Moore, 1991/2014).

So consensus reverse iteration is not Rogers or Moore played backward. It does not ask the late majority to be first to buy an unfamiliar technology, and it does not turn the same batch of users, step by step, into innovators. What is reversed is the order in which a product team builds the product and bears uncertainty: survive first on a mature problem, then widen capability and the radius of innovation on the evidence.

4. Duan Yongping and BBK: after coming later, still have to lead #

Keep only the first half of "dare to be last" and the method shrinks into waiting and copying. BBK's VCD years show that a later firm still pays the full cost of competition.

By 1997, when BBK was putting VCD advertisements on air, the Chinese market was already crowded. Brands such as Shinco, Aido, and Malata held the main positions. In February 1998, Guangming Daily, citing Sino Market Research, had Shinco, Aido, and Malata together already above 60%, and BBK not yet in the listed top ten (Guangming Daily, 1998). That demand had already been proved does not mean a later firm had already obtained a place.

Chapter 7 already recorded BBK's 1997 entry into VCD and a quality grading. What belongs here is what competitive cost was paid after coming later. A company chronology held by CCTV also shows that in November the production quality-assurance system passed a production-license review. Another company document, citing Sino data, says that in 1999–2000 BBK's disc-player market share sat in second place. Those materials include publicity supplied by the company. They cannot be used to estimate a typical effect of the method. They are enough to see that a later firm still paid the full cost of quality, brand, channel, and service.

What a mature category reduced for BBK was one layer of uncertainty: whether VCD already had market demand. It did not answer for the company what product to make, how to hold quality, under what channel conditions continuing sales would form, or how after-sales would cover the country. It also did not guarantee that a place would still be left after Aido and Shinco. Dare to be last settles which class of problem to enter. Later-but-better decides whether a later firm has a right to remain.

Chapter 7 already cited the 2025 exchange at Zhejiang University: once scale reaches a certain point, and no one sits ahead, new things also have to be done; coming later does not guarantee success. Starting from a mature consensus can first use a more certain problem to accumulate cash, data, the capacity to deliver, and judgment. After the organization's position has changed, a path that was reasonable then has to be judged again.

5. One main path, five stages #

Consensus reverse iteration keeps only one main path:

Mature consensus → survival product → complete product → advantage product → frontier product

Figure: each stage uses results to decide keep, upgrade, roll back, or stop. A product can remain at any stop for a long time. The frontier is not a diploma.

Mature consensus: first prove that the problem does not depend on your technical story #

Mature consensus is not a class of conservative users. It is a set of evidence about a problem. Users are already paying for an old solution, continuously spending labor or time, or bearing a clear loss if the problem is not handled. Even if the words AI, robot, or new material are taken off, they still know why the matter has to be dealt with.

The value this stage has to verify is whether a new method can improve an existing result without first requiring the user to understand the underlying technology. Valid evidence includes the target scene, old spend or substitute behavior, a reachable channel, and an improvement in result under real conditions. How new the technology is does not count as evidence of demand.

If there are only statements in interviews and no continuing behavior, or if the problem is real but there is no budget and no channel that can be served, the organization can stop at mature consensus. Continuing a scaled outlay then books an R&D project that has not yet passed payment and delivery as if it were a business.

Survival product: first verify payment, delivery, and unit economics #

A survival product is not a cheap edition, and it is not finished when the first payment arrives. It has to prove three things at once: the target customer has already paid for a smallest result that is still complete; the team can deliver what was promised; and adding one more customer does not send support, service, and customization cost into a limitless worsening.

Evidence for entering the next stage includes repeat payment or continuing use, a delivery cycle that can be recorded, a way of implementing that is becoming stable, and unit economics that hold, or can credibly be improved, at realistic volume. Those results separate a product from a demo. If every order still needs the founder to put out a fire, and the more revenue there is the larger the loss and the complexity, that is still a set of projects. It is not a survival product.

A survival product can also be a long-term end. A product of modest scale, stable delivery, and healthy profit does not have to enlarge uncertainty merely to add technical novelty.

Complete product: let a typical customer obtain the result on their own #

That the core technology works does not mean the product is complete. The customer still has to finish buying, going live, migrating, integrating, learning, getting support, and recovering from a failure. Moore called the full set of conditions a mainstream customer needs the "whole product." In the five-stage path, complete product tests whether delivery has left a handful of heroes. It does not make another ladder of adoption.

The value a complete product has to verify is whether a typical customer in the same target scene can get the result steadily. Evidence for entering the next stage can come from retention or repurchase across several batches of similar customers, repeatable implementation, a falling burden of support, a clear service level, and customers who can be cited. Adding many features, while sales still promise and engineers still patch by hand, is not completeness.

A complete product can likewise hold for a long time. Reliability, ease of use, and maintainability are value. They are not a junior stage on the way to the frontier.

Advantage product: prove that the difference can be repeated #

An advantage product does not only replace an old method. It has to bring, steadily, a result the old method could not easily supply: a lower total cost, a shorter cycle, higher quality, less risk, or a stronger capacity to earn. It has to exclude three false advantages: orders obtained only by cutting price, customers obtained only by relationship, and a polished case customized for a single large customer.

Before the next stage, it should be possible to compare the new and the old on the same definition, to see a customer result that repeats, a reasonable price or a lower total cost, and a channel and a way of delivering that can extend. If the advantage disappears as soon as low price, relationship, or a one-customer customization changes, the difference cannot yet be repeated.

If the difference is already enough to support a long operation, an advantage product has no need to keep walking toward the frontier. The radius of innovation is set by customer results and operating evidence, not by how new the technology looks.

Frontier product: put a new capability into a bounded exploration #

A frontier product faces a new use that is not yet mature. Valid early signals include: a set of customers has in fact met a task that used to be hard to finish, has already put budget, data, or people into a joint trial, and can supply feedback fast enough.

The value this stage has to verify is whether the new capability has opened a use worth continued investment. The result of an exploration can support only a new problem or a new method that may hold. It does not equal a mass market already formed. Real behavior, joint investment, technical feasibility, a bearable budget, and a clear stop line are the basis for continuing.

Not every product has to enter a frontier stage, and more frontier is not more advanced. A complete product on a mature problem can create more value than a frontier product that still lacks evidence of use and payment. Where a product stops is decided together by mission, customer value, limiting resource, and risk.

The public method of 与爱为舞 (Dance with Love) supplies a limited mechanism on the side. In a transcript published by Qiming Venture Partners, founder Zhang Huaiting put the company's path as: first close a business loop and verify that the scene works, then let the model gradually assist or replace steps inside it, and finally complete the AI change of the business (Qiming Venture Partners, 2025). That sits near the five-stage order of building. It cannot prove that the five stages work in general. The company's financing, revenue, users, and teaching results are also mainly company figures. Chapter 14 opens the case at the layer of organizational assembly.

6. R&D has to walk with the evidence #

The product path decides where to begin. R&D discipline decides whether the organization can learn from each round of results.

The first discipline is: verify one main uncertainty at a time, as far as possible. A main uncertainty is the unknown that most affects the next decision. The team can first write what this round actually has to judge—price, frequency of use, way of delivering, or technical feasibility—and then keep the other conditions as stable as it can. The aim is to raise attributability. It is not a claim that a complex product can forever change only one variable.

A strict "one factor at a time" misleads when factors interact. NIST's handbook of experiment design notes that if the effect of one factor changes with the value of another, changing factors one by one will not see that interaction. A complex product has to use, according to the problem, staged release, a controlled trial, a multivariable design, or monitoring of interactions (NIST/SEMATECH, 2012). Changing a new price, a new interface, and a new segment of customers at once makes explanation harder. It does not send the value of learning automatically to zero. The right move is to admit the mix, improve the experiment, and not invent a certain answer.

The second discipline is to separate stable core, adjacent improvement, and frontier exploration.

The stable core is the present source of payment and of value already verified. R&D has to give it rails of reliability, safety, compatibility, migration, and rollback. Adjacent improvement adds only one layer of capability or scene, so that the new result can still be compared with the present core. Frontier exploration is isolated from the stable core, with budget, term, technical evidence, market evidence, and a stop condition set in advance. A survival period can keep exploration inside a cap. It cannot let a failed exploration pull the core down.

The third discipline is: a new capability must not break old value by accident. When a new capability changes a verified flow, present customers bear the cost of learning, migrating, failing, and rolling back. Before a release, write compatibility, cost of migration, a rollback plan, and the rail metrics. When an old path in fact no longer creates value, it can be withdrawn. Withdrawal itself has to have evidence. A function that still has use and payment cannot be booked straight as technical debt.

The fourth discipline is: results decide the next step. A vision does not sign for the results. A vision can propose a hypothesis. A roadmap has to allow keep, upgrade, roll back, and stop. The last round of evidence answers only the question written in advance. The organization still has to put customer behavior, burden of delivery, economics, and risk together before it decides whether to widen the radius of innovation. There is no universal timetable on which "three signals all light," and there is no need to wait for adjacent users to walk over on their own before admitting that the next layer of demand exists.

The fifth discipline is: settle a standard each round. When an iteration ends, at least record the hypothesis, the conditions of use, the action taken, the result observed, the exception that appeared, the reason for stopping or continuing, and what the next round has to revise. Code tells the product what it can do now. A standard tells the organization how to judge the next time a similar problem appears.

The five disciplines together replace the urge to showmanship. R&D is not stuffing the most frontier capability into the product as fast as possible. It is, without breaking value already verified, continuing to buy more reliable judgment.

7. Mature demand can also hold a company in place #

Starting from a mature consensus only reduces part of the uncertainty of demand. It also produces six costs.

The first is the technology window. A key standard, supply, license, intellectual property, or place in an ecosystem may fix itself quickly. If a later firm would then lose the right to choose for good, slowly accumulating capability from mature demand may come too late. A reverse path has to constrain its pace with a real window. "Do it later" cannot be treated as an unlimited wait.

The second is network effects. The value a user obtains from the same network, from compatible products, or from complementary software or services may rise as more people take part (Katz & Shapiro, 1985). A network effect is not, by itself, winner-take-all. Whether it forms a barrier to entry still depends on its strength, switching cost, compatibility, and whether users can use several platforms at once (OECD, 2022). Asking only "is this social, or a platform?" is not enough to decide that a firm must run forward first.

The third is channel cost. Mature users may not gather in technology media and product communities. They may depend on agents, offline stores, industry relationships, and long service. BBK's later rise also sat on national channels, advertising, and after-sales outlay. Educating an early market can be expensive. Reaching and serving a mature market can be more expensive. The two ledgers have to be compared in the concrete scene. Neither side can be declared cheapest in advance.

The fourth is price sensitivity. When a deal can be closed only by continuing to cut price, the cost the technology saved is easily eaten by channel, implementation, returns, and support. A survival product has to count the full cost of delivery, not use the first payment as proof of a business model. Price can be low. Future maintenance cannot be hidden inside today's product margin.

The fifth is customization debt. A special change for a large customer may buy revenue and important learning. If every customer keeps a special logic, testing, upgrades, and support grow with sales. The team has to separate configurable, reusable product capability from custom projects that should be priced and staffed under a service contract. A special feature with no owner, no maintenance budget, and no condition for deletion will slowly turn a survival product into a projects firm.

The sixth is technical debt. Delivering before cash runs out often means accepting a local compromise. The problem is not that the debt exists. It is whether the organization has recorded what it affects, who is responsible, and when it must be repaid. If gaps in data quality, safety, and architecture begin to hinder the stable core, further expansion only enlarges the cost of the repair.

Those risks also show that a reverse path is not a generally superior solution. Tesla's 2006 Master Plan chose, in public, to start from a high-priced Roadster and then, with later models, raise volume and lower price. By the end of 2012 the company disclosed about 2,450 Roadsters delivered and about 2,650 Model S cars (Tesla, 2006; Tesla, 2013). That at least shows that opening from frontier technology and from users with a high willingness to pay can move, step by step, toward a larger market. It cannot prove that a forward path is naturally superior, and it cannot book the whole education of the electric-car market to Tesla. That a forward path can hold depends, likewise, on product, capital, infrastructure, policy, supply chain, and long execution.

The choice of path has to follow what the organization can currently bear, and has to locate again when the results change.

8. Meta: missing consensus is only one item #

The public record of Horizon Worlds presents three verifications that were not finished: mainstream demand, product quality, and the product's own constraint of survival. Profits from Meta's other businesses let a large outlay continue.

Internal files seen by The Wall Street Journal in 2022 showed that Meta first set Horizon Worlds' year-end 2022 monthly-active target at 500,000, later cut it to 280,000, and that actual monthly actives were then under 200,000. Most visitors did not come back after the first month, and the user count had been falling since spring. Of the worlds creators had built, only 9% had ever been visited by at least fifty people, and most had never been visited (The Wall Street Journal, 2022). Another "300,000" often mixed into this set is the combined monthly actives of Horizon Worlds and Horizon Venues in February 2022, confirmed to the press by a Meta spokesperson. It is not a peak for Horizon Worlds alone.

Weak evidence of demand was not the only problem. An internal memo dated September 15, 2022, put the product under a "quality lock," to repair gaps in quality, performance, and stability first. Vishal Shah, who led the product, wrote that employees' own usage dashboards showed that many people on product teams rarely entered Horizon. A follow-up memo on September 30 was still asking teams to use the product each week. The material also named onboarding problems in the first-use flow, and said plainly that the product had not yet found product-market fit (The Verge, 2022).

What Meta faced, then, was a set of problems acting together: the friction of entering with a headset and controllers, a social experience weakened by empty worlds, density of content and retention, product quality and onboarding, and whether "socializing in a virtual world" could enter enough people's ordinary days. Booking the whole result to "users had no consensus" misses the part in which the product itself did not keep its promise. Writing it as if no scene of use existed at all also erases VR communities and creators already there.

Reality Labs' losses likewise cannot be booked straight to Horizon Worlds. In segment data Meta disclosed to the SEC, Reality Labs' operating losses from 2020 through 2025 were $6.623 billion, $10.193 billion, $13.717 billion, $16.120 billion, $17.729 billion, and $19.193 billion, $83.575 billion in all. The segment covers virtual- and augmented-reality hardware, software, and content together. The 2025 annual report also lists Quest, Horizon, wearables, AI/AR devices, neural interfaces, and long-term underlying technology. $83.575 billion can show that Meta bore a large long-term outlay for a next computing platform. It cannot show that Horizon Worlds alone burned that sum (Meta, 2022/2024/2026).

The product changes of 2026 also have to be written accurately. Meta first announced in March that Horizon Worlds would turn toward mobile and was planned to be removed from Quest after June 15, then quickly changed the arrangement for a full shutdown. Later official developer text said that individual Worlds would no longer occupy a listing in the Quest store, that existing Worlds could still be reached through the Horizon Worlds app, and that almost all new emphasis for Worlds would turn to mobile, while the company would keep investing in Quest, the VR platform, and third-party VR developers (Meta, 2026). It cannot be written as if Quest Horizon Worlds had already been fully taken down on June 15, 2026, and still less as if Meta had exited the headset business.

BBK borrowed a category demand that had already shown itself, and still had to compete for a place with quality, channel, and service. Meta had the capital to bear a frontier exploration. Capital cannot replace product-market fit. The former cannot prove that coming later always wins. The latter cannot prove that going first always loses. The standard of judgment that follows is this: first list the uncertainties the product has not yet verified, then check whether the company can bear them.

9. The frontier is not a diploma #

Write the five stages as "do the mature market first, then walk all the way to the frontier," and the frontier is set again as the top of a value rank. The path is also mistaken for a promotion table that has to be cleared.

The five stages have only one building discipline: choose the path on present evidence, then let the next round of results decide the radius of innovation. Mature consensus can be a start. A survival product can be an end. A complete product and an advantage product can be operated for a long time. A frontier product can also stop after verification fails. What is accepted is not whether the product stands at the front. It is whether the limiting resource has entered a direction nobody needs, that cannot be delivered, or that cannot be borne.

Originality has to be tested by customer value. A vision has to be tested by evidence of the stage. Frontier exploration may not occupy, without a cap, the survival capital of the stable core. Those three constraints together replace reliance on originality and the practice of letting a technical display stand in for a result.

When the product team and the customer sit at nearby stages, demand, evidence, and delivery are easier to align. When the team has already moved to the next stage and the customer is still protecting old value, or when the maturity the customer requires sits far above the team's present capability, a product problem turns into a relationship problem: keep compatible, screen again, or exit? That mismatch is what the next chapter takes up.

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

Pick one product that is being opened, that is about to receive a larger outlay, or that is about to be rebuilt. Do not audit the whole company at once.

1. Answer the two questions of the line of work. Write one sentence for "why now" and one for "why it will still be worth doing in ten years," and attach one piece of observable evidence to each. Where evidence cannot be written, mark a hypothesis. Do not enter a scaled outlay.
2. Locate the target demand. Do not pin innovator or conservative labels on the user. Write how they currently handle the problem, what money, time, labor, or risk they have already spent, and whether the problem still holds after the technical nouns are taken off.
3. Locate yourself. List the runway of usable capital, the real cost of channel, the technical capability that can currently be delivered reliably, and the upper bound sales, implementation, and support can bear.
4. Choose a path and a present stage. Say why, at this moment, a forward path or consensus reverse iteration fits better. If the reverse path is taken, mark only one present position on "mature consensus—survival—complete—advantage—frontier," and write the value already verified.
5. Freeze the next main uncertainty. For the next round, write the question, the metric, the rails, the budget, the term, rollback, and the stop condition. Then mark which work belongs to stable core, adjacent improvement, and frontier exploration.
6. Write the result back into a standard. When the iteration ends, record not only what shipped, but which judgment was supported, which was overturned, how the conditions of use changed, and whether the next round will keep, upgrade, roll back, or stop.

A week later, do not count how many stops the team still sits from the frontier. Check three results: whether the reason for payment is clearer, whether delivery is more repeatable, and whether the next piece of limiting resource already has evidence for continued investment.

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

1. Claim restatable in one sentence ✓: locate demand, locate self, then match; when conditions fit, start from a mature problem consensus, verify payment, delivery, and survival, then widen capability and the radius of innovation on results.
2. Nine-section spine and whiteboard figure ✓: where judgment goes → two questions that cannot be multiplied → locate demand then self → BBK later-but-better → five stages → R&D with evidence → six costs of mature demand / Tesla as forward boundary → Meta as more than missing consensus → frontier is not a diploma. Figure is the five-stage path plus R&D rails; normal-distribution six-tier ladder is out.
3. Evidence caliber ✓: Duan and BBK from Chapter 7 plus Guangming Daily 1998 Sino top-ten, 1997 license, 1999–2000 second-place as company-supplied. Horizon 500,000 / 280,000 / under 200,000, 9% of worlds, combined 300,000 as Worlds+Venues, quality lock and PMF from WSJ 2022 and The Verge 2022. Reality Labs $83.575 billion as the segment, not Horizon Worlds alone. 2026 mobile shift, not a Quest takedown and not an exit from headsets. Tesla 2006 Master Plan and ~2,450 Roadsters / ~2,650 Model S by end-2012. 与爱为舞 as Qiming 2025 mechanism only. bioby.ai as judgment at choice. NIST 2012 on interaction. Katz & Shapiro 1985; OECD 2022. Amazon 2009 as in Chapter 6.
4. Case contract ✓: BBK and Meta as a mechanism pair, not a same-industry win-loss experiment. Reverse path not a generally superior solution. Gaokao / HR / six-tier chasm, variable×invariant, reverse adoption curve, Interference Method isomorphism, Qudian / Chegg, OPPO half the phone market, 1996 BBK entry, Meta headset exit, Dance with Love operating numbers, Duan's Tesla miss, career note, and three-signal timetable are out.
5. Path and R&D ✓: one main path of five stages, each with a stop; one main uncertainty not one-factor-forever; stable core / adjacent / frontier; no three-light ritual.
6. Chapter boundary ✓: full 与爱为舞 ledger waits for Chapter 14; relationship mismatch waits for Chapter 13; Ch7 holds 1997 entry and quality rating.
7. Fluency ✓: rewritten in English voice from the Chinese authority; current prose-standard.