Chapter 6: Standards: Stop Execution When No Standard Exists

1. Two eras of a bowl of noodles #

Liuzhou people have been making snail noodles for decades. There is a consensus on which soup is delicious, how smelly the sour bamboo shoots should be, and how chewy the rice noodles should be. But this consensus only lives on the hands of the chef and the tongues of the diners. The consensus that lives there has a fatal property: it cannot be replicated. It takes three years for a master to teach his disciples. Once they leave Liuzhou, they will lose their flavor, and they will spoil once they are put in bags. Therefore, in the past few decades, the industrial form of this bowl of noodles has been: nearly 5,000 physical stores across the country, each based on the feel.

In June 2015, the Liuzhou Municipal Government established a leading group for the preparation of local standards; in May 2016, the "Local Standard for Food Safety Liuzhou Snail Noodles" was officially implemented: materials, packaging, physical and chemical indicators, microbial indicators, a bowl of noodles were written down into testable provisions one by one. There is a detail in this set of regulations that deserves to be framed by all those who write standards: the moisture content of rice noodles is set at 14%. This number is not taken from a photo: high moisture content breeds mold, and low moisture content makes it easy to break; it was frozen into the standard after repeated tests, sending samples for testing, and listening to the company's opinions on production capacity. **Every number in the standard should be a verified judgment, not a professional number. **

Then the industry curve started. The number of bagged snail noodle companies increased from 1 in 2014 to 113 in 2020; bagged sales revenue in 2020 was 10.994 billion yuan, a year-on-year increase of 75.74%, and the total export value was 35 times that of the previous year; in 2021, the sales revenue of the entire industry chain was 50.16 billion yuan, and the annual bagged delivery volume exceeded 100 million pieces. Please note the cause and effect in the chronological order: the flavor consensus existed for decades, and the curve remained flat; the moment the consensus was written into the standard, the curve began to steepen. Because from that moment on, the judgment of this bowl of noodles (what counts as qualified) can leave the hands of the master for the first time and be executed by the factory's production line, quality inspection instruments, and manufacturers on the other side of the ocean.

That's what this chapter is about. The word standards has a reputation in most companies that borders on bureaucracy (documents, stamps, processes). This chapter aims to find its true identity: ** standards are not bureaucratic documents, but the way judgments are stored. **

2. Standard triple identity #

Putting the standard into the framework of this book, it has three identities, each more valuable than the last.

The first level: the solidified substance of judgment. Chapter 4 says that judgment is an organizational asset, but gaseous judgment (the intuition that lives in someone's head) is not an asset, but a risk exposure (Wang An Computer demonstrated it with bankruptcy). Only when judgment is refined into solid provisions of "under what conditions, under what trade-offs, and where is the qualification line drawn?" can the transformation from personal capabilities to organizational assets be completed. The 14% of snail noodles is a phase change: the sense of propriety in the hands of the master has become a number that any quality inspector can implement.

The second level: the organization’s compound interest container. A verified standard will recover the verification cost when used for the first time, and will be a net benefit every time it is used thereafter, and users do not need to be smart again. Chapter 3 says that individuals rely on adaptive insight to borrow standards, and Chapter 4 says that organizations rely on standard libraries to avoid repeated groping; this chapter adds the nature of the container itself: standards are the only asset in the organization that get cheaper the more they are used. People get tired, judgments fluctuate, standards don’t.

Third level: AI command interface. This is the new identity that crowns standards in the AI ​​era, and it is also the second half of the conclusion of this chapter. All interactions between you and the AI ​​(prompts, evaluation rules, context files, workflow definitions) are essentially standards: you are telling it what counts, what not to do, and what to order. This means a cruel conversion: The standard quality you give the AI ​​is the upper limit of the AI's output. Klarna in Chapter 2 has already demonstrated the opposite: feed the AI ​​a wrong standard of "cost first", and the AI ​​amplifies it with perfect execution. In the old days, standards were vaguely written, and human employees would use common sense to quietly fix them. Bad standards could run sick for many years; AI doesn't fix it, it only amplifies it. SOP used to be a management document written for humans to read, but now it is an instruction issued directly to the machine. **In an era when execution is free, standards are your only remaining control over execution. **

3. Rules of this chapter: If there is no standard, execution will be stopped. #

After the triple identity is established, the rules will naturally emerge. Since execution is free and judgment is scarce, the most dangerous situation for something is to be executed at high speed without standards. Free execution multiplied by absent judgment equals manufacturing rework at maximum efficiency. Chapter 3 has already demonstrated why at the individual level (the feeling of non-standard execution is exactly enrichment, and busyness provides the illusion of growth). This chapter will not repeat it, but only gives the organizational version of the operational provisions:**When you find that there is no standard for something, immediately stop continuing to implement it; first find, borrow or establish standards, and then resume action. **

Three operational issues, explained one by one:

Who has the right to stop? Any discoverer. The most easily distorted aspect of this rule is that the right to suspend is made into a rank privilege, which is equivalent to adding another layer of approval on top of unstandardized execution. The correct design is to make it an obligation: anyone who finds that he or his team is performing without standards will report it and call it a day. The manager's responsibility is not to approve the stop, but to respond to the stop: the organization borrows standards or establishes standards within the agreed time limit. An organization that cannot stop does not have strong execution capabilities, but has failed brakes.

Stop for what? The order is fixed: borrow first (who has done it before? The standards and applicable boundaries are borrowed together), then test (do the migration conditions hold in our scenario?), and finally establish (cannot borrow or test, use minimum cost experiments to create new standards). This order cannot be messed up. To establish first and then borrow is to pay twice, and to borrow without verifying is the way Johnson died in Chapter 3. The complete action chain unfolds in Chapter 7, and this chapter only determines the order.

When will it be restored? The answer is with enforceable standards. Pay attention to the acceptance criteria of the word "executable": newcomers (or AI) can run the same level of results according to it. What doesn't meet this caliber is the impression, not the standard.

Stopping is not procrastinating. In the era when execution is expensive, the cost of stopping is real money; in the era when execution is free, stopping to look for standards is the lowest-cost execution action**. What you lose is a false sense of progress for a few days, and what you get in exchange is the freedom of not having to finish the mistakes.

Give this iron law an extremely small example, and use its smallness to illustrate what "default behavior" means. In June 2026, our Shenzhen office will be relocated. I want to turn a 15-square-meter room in the new office into a conference room. The budget is 10,000 yuan. The goal is to experience the comfort of a single sofa chair and the formal feeling of receiving guests. Before starting construction, find the standards: How can these two experiences be established at the same time in 15 square meters? Who has done it? What to check for acceptance? After searching for three days, I couldn't find a mature answer that I could borrow. So pause. As of this writing, that room is still empty. To decide whether a project worth RMB 10,000 is worth treating like this, let’s spread the accounts and calculate: the cost of stopping is three days of searching and a temporarily vacant room; the cost of not stopping is spending RMB 10,000 on a plan that can’t even be said to be “done”, and there is a high probability that it will be spent again. And the real reason lies outside the ledger. Chapter 2 defines organizational DNA as a default behavior that occurs automatically when no one is supervising: Position 1 can only stop at small things worth tens of thousands of dollars, and "stop executing if there are no standards" will grow into the muscles of the organization; usually "do it first and talk about it later" on small things, but when it comes to decisions involving tens of millions of dollars, everyone is suddenly asked to stop and look for standards, and no one will take it seriously. **Small things are the practice ground of genes, big things are just its examination room. **

4. standards engine: circulation pump and ladder #

The framework diagram of this chapter is a machine: Judgment (gaseous state) → Refining → Standard (solid-state document) → Delivery (to humans and AI) → Execution → Data reflow → Standard iteration. A circulation pump sucks in the judgments that happen every day in the organization, and spits out an increasingly accurate standard library at the outlet. The comparison picture of straight lines and bifurcations next to it has been drawn at the personal level in Chapter 3. Here is just one annotation: the same picture, replace "one person's path" with "a thousand people's path", and multiply the cost of bifurcation by one thousand (as shown in Chapter 3, no re-argument).

Circulation pump answers "How to produce standards", but there is still one question left: What standards does the organization need? The answer is an eight-story staircase:

1. MissionAnswer why it exists: Use AI to improve all humankind.
2. BeliefAnswer what you believe in the long term: Human-machine collaboration is the future of mankind.
3. Standards of Conduct constrain how to act: Treat People with Honesty.
4. Thinking standards stipulate how conclusions are produced: borrow first and create later, two-sided samples, medians for statistics, first principles, and the book's Clarity Method (expanding one's own options) and Interference Method (narrowing the other party's options, Chapters 7 and 8 each use a whole chapter to expand). These thinking frameworks all live on this level.
5. Judgement Standard is responsible for how to choose: Judge by Value.
6. Personnel investment standard tests whether human investment can form a positive cycle: the ROI of all employees is greater than 1, and the verification window changes with the stage.
7. Management Standards are responsible for how to allocate decision-making authority: Instruction → Guidance → Negotiation → Delegation → Delegation.
8. Learning Standard is responsible for how to correct the system: reflection (examination of myself) + Feynman technique.

Let me first explain the origin of these eight levels of sample provisions: they are not fictitious templates for this book, but the organizational standard matrix that my own company, bioby.ai, runs every day. The sentence behind each layer is the actual implementation of our employees and AI. Printing the original text of our own operating system into the book means that this set of standards is subject to two tests at the same time: your scrutiny and our own operating results. The complete assembly process (in what order and how many weeks it took to erect the eight floors) is shown in Chapter 13 with a full case study of this company.The eighth floor is not a wall of juxtaposed slogans, but a chain of cause and effect: mission determines direction, belief determines long-term assumptions, behavioral standards constrain means, thinking standards stipulate what to think, judgment standards are responsible for choices, personnel input standards test whether the value can cover limited resources, management standards allocate decision-making rights, and learning standards continuously modify everything in front. Chapter 4 has broken down the division of labor between thinking standards and judgment standards (process standards govern the process, acceptance standards govern the results), and here we look at their positions in the entire chain: how a conclusion is produced, how options are sorted, and what is separated in between is the dividing line between "thinking" and "selecting".

Two instructions for use. First, the stairs are not built in one go: only the panorama is given in this chapter, and the sequence and rhythm of assembly are left to Chapter 13. Second, the three words values ​​do not occupy a separate layer on the ladder, because values ​​do not enter the judgment, management and learning process, and are just adjectives hanging on the wall. The lower five levels of the ladder are the execution form of values.

5. What is worth writing into a standard: laying the foundation on invariants #

Before starting the standards engine, there is another site selection problem: judgments occur in hundreds or thousands every day, which ones should be solidified first? A standards engine that writes the wrong object will efficiently produce useless documents.

The principles of site selection have their origins. Bezos has been asked "What will change in the next ten years" for thirty years, and his answer is: This is the wrong question. "I almost never get asked, 'What won't change in the next ten years' - and the second question is actually more important, because you can build your strategy on things that are stable over time." There are three constants in Amazon's retail business: customers want lower prices, faster delivery, and more selection. "I can't imagine that ten years from now, a customer will say: I love Amazon, but I just wish the prices were higher and the delivery was slower." He also added a less quoted but equally sharp saying: "This question does not require a lot of research - these things are big and fundamental, you know it yourself."

Translated into the language of this chapter: Standards should be built on invariants first. Standards built on invariants do not need to be rewritten frequently. The verification cost is recovered once and compounded for many years. Amazon’s three constants have governed resource allocation for twenty-five years (the shareholder letter in Chapter 5 is their capital version). To determine which of your invariants are yours, just use Bezos’s caliber to ask: You can’t imagine customers asking for the opposite of it ten years later. What will happen to the standard based on variables depends on a sample.

6. Loser sample: companies that are cleared every two years #

Qudian Group was listed on the New York Stock Exchange in October 2017, and its market value once exceeded US$10 billion. The campus loan and cash loan business gave it the first pot of gold, and also gave it the first set of proven capabilities: online customer acquisition, risk control model, and operation system for loan repayment.

Then regulations tightened, the main business came under pressure, and Qudian began to change its track. In 2018, it started Dabai Auto (new car retail) and stopped selling in May 2019; in 2020, it started Wanlimu (luxury e-commerce) and Wanlimu Children's (education), each of which died in about a year; in 2022, it transformed into prefabricated dishes and got off to a high-profile start. In the same year, it received two US$1 delisting warnings from the New York Stock Exchange. The market capitalization was approximately $290 million, approximately 97% evaporated from its peak.

Using the framework of this chapter to do an anatomy, Qudian’s problem is not that it “selected the wrong outlet” (each of the air outlets it chose is not outrageous on its own), but that every time it changes the air outlets, it discards all the standards settled in the previous process and starts from scratch. None of the customer acquisition standards, risk control standards, and operational standards have been asked "which parts are built on invariants and can be migrated to new scenarios." Comparing the four actions in Chapter 3, this is the organizational absence of "judging migration conditions": it is not that the wrong standards are borrowed, but that there are no standards at all. Compare Bezos: Amazon’s three invariants haven’t changed in twenty-five years, while Qudian’s scoreboard changes every two years. **The outlet will change, but each outlet has its own scoreboard; the standard is based on the outlet variable, which is equivalent to clearing the organizational judgment every two years. **

Qudian and snail noodles are particularly dazzling when seen together: a bowl of street snacks, because it has written the consensus into a standard, has grown from a company into a 50 billion industry; a tens of billions of dollars company, because it has never solidified the standard, has transformed back to the starting point four times. The asset is never the business itself, but the transferable standards accumulated by the business. **Losing standards is much more expensive than losing business, but the bills come slowly. **

7. Standard standards: Why is the median an iron law? #

This chapter gives a complete example of the thinking standard layer in the eight-level ladder, and also answers a question that smart readers will ask: "What standard is used to write the standard itself? Will it be infinite recursion?"

An example is that iron law of statistics that recurs throughout this book: Report the median, not the mean. The reason is not in mathematics textbooks, but in organizational behavior: averages will be hijacked by outliers, and organizational decision-making is most likely to be hijacked by star cases. A sales team's per capita performance may be 800,000 yuan. It may be ten people who all earn around 800,000 yuan, or it may be one star with 5 million yuan and nine with 300,000 yuan. The average would give the same number for both situations, and the median would honestly quote a world of difference between 800,000 and 300,000. If you use the average to make personnel decisions, you will set goals for all employees based on the phantom of the star; if you use the median, what you see is the true level of the organization. **The average lies because it is afraid of outliers; the median is honest because it only recognizes the majority. **This rule answers the recursive question: the standard of standards is the standard layer of thinking. It stipulates all standard production processes (how to collect evidence, how to count, how to verify), and it is very small in number and extremely stable (borrow first and then create, bilateral samples, median, you can count them all on one hand), and the recursion ends at this level. The foundation does not require another foundation.

8. Boundaries of Judgment #

Three articles to prevent this conclusion from being used as a new form of bureaucracy.

First, not all judgments are worth solidifying. The production capacity of the standards engine is also a limited resource (the discipline of Chapter 5 applies here): only judgments with high frequency and high error costs are solidified, such as the most reworked processes, recurring trade-offs, and links taken over by AI. Low-frequency, low-cost, highly situational judgments leave people with on-the-spot judgments, and the costs of forcibly standardizing them are higher than the benefits. The health indicator of the standard library is not the number of items, but the reference rate of each item. A standard that no one quotes is a waste of paper in uniform.

Second, the standards will become outdated and the rules themselves have amendment provisions. The last level of the eight-level ladder (learning standards) is specifically responsible for revising the previous seven levels; Bezos's "invariants" are not "everything remains the same": in addition to the three invariants, Amazon's specific play style has changed countless rounds. The lesson learned from Qudian is not "not allowed to change directions", but rather to not conduct a standard migration audit when changing directions: which standards are built on invariants and should be taken away, and which standards are built on old scene variables and should be left behind. It's okay to change directions, but not to run naked.

Third, tell the truth about the quality of the sample. Snail noodles are a government-led industry standard sample, which is not the same as the internal standards engine of the company. This chapter only adopts the mechanism that "consensus can be replicated on a large scale only after it is solidified into a standard" and does not attribute all the credit for industrial policies to the standard (industrial parks, logistics, and e-commerce dividends are all included in the curve). The instance of internal triggering "Stop Execution" (, the conference room in Section 3) has been added. It is my own self-report. Please listen to it at a discount. Another open space for honest marking: a company with SOP culture turned SOP into an enterprise-level case of AI interface. It has not been completed yet, and I am looking for it. The load-bearing structure discussed in this chapter is the derivation of the triple identity of the positive and negative contrast between Jiaqu Store and Snail Noodles. After filling up the empty space, it will be more three-dimensional, but it will not affect the load-bearing.

What to Do Monday Morning (No. 1 perspective) #

Three steps, within an hour:

1. Find the focus: Choose the type of job that has the most rework in the company and ask a question: Does it have a written qualification standard? If not, I will write it today; if I cannot write it, it means that the judgment of this type of work has not been refined yet. This is a topic that is truly worthy of a meeting, and is ten times more valuable than a rework review meeting.
2. AI Touchstone: Feed the written standards directly to the AI, and let it produce a version according to the standards. The quality of AI's output is a mirror of the quality of your standards. Human employees will use common sense to quietly fix vague standards, but AI will not. It will magnify every ambiguity for you to see. Change the standards and feed again until the AI's output makes you say "passed". This standard has only now been written.
3. Establish rules: Announce the stop rule to the management: Anyone who finds that there is no standard implementation will report it and stop it, and the manager will respond within the time limit. At the same time, the supporting terms were announced: no accountability if stopped, only accountability if the situation is kept secret. You will most likely receive a lot of calls to stop in the first week. Don’t panic. It’s not that the chaos has increased, but that the existing chaos is visible for the first time.

Side note (individual vs. team perspective): The AI ​​litmus test holds true for individuals as well. You are always dissatisfied with letting AI work. Don’t blame the model first. Read your prompt as the standard you wrote: How many testable qualifying conditions are there? Most people will find themselves giving the AI ​​wishes, not standards.

Chapter Acceptance Self-Check (compare with the five acceptance standards of the chapter)1. A sentence of assertion can be restated ✓, and it is an inference of the core assertion (adaptive insight needs to be written into a standard before it can be reused → no standard execution equals zero judgment execution). #

2. Whiteboard frame diagram ✓ (standards engine circulation pump + eight-layer standard ladder; straight line/bifurcation diagram refers back to Chapter 3 without repeating the argument).
3. External comparison and data ✓: Positive branch Liuzhou Snail Noodles (government documents and industry data have been verified, caliber boundaries have been explained) + S3 Bezos invariant (HBR 2007 + re:MARS 2019 primary sources have been verified); loser Qudian (financial report level data has been verified); The conference room example has been written into Section 3 (2026-07-26, the quality of the self-description has been stated in the thesis boundary); the "SOP→AI Interface" space has been truthfully marked.
4. 12 candidates for golden sentences ✓ (The sentence "straight line/bifurcation" belonging to Chapter 3 has been removed, and "practice field/examination room" has been added).
5. "What to Do Monday Morning" Three steps from the No. 1 perspective + personal notes ✓.