The Reconciled Creed

Twelve isms went into this book. Fourteen come out, and eleven of the original twelve are narrower than they were.

Below, each one appears twice: as I have been saying it for years, and as it stands after being tested. Where the second version is longer, that is not hedging. Every added clause is a condition the original was silently assuming, and the assumption is what failed.

I want to say one thing about the shape of the list before giving it. I expected, starting this, that a few isms would survive intact, a few would need a qualifier, and one or two might have to go. What actually happened is that none were abandoned and almost all were narrowed, and the narrowings turned out to be of three kinds.

Some added a domain: where the rule holds and where it inverts. Some added a cost: what has to be paid, permanently, to keep the rule true. And some added a person: who the rule acts on and what standing they have. The first kind I half expected. The second kind is the one that changes how any of this gets implemented. The third kind is the one that was not in the creed at all and is the reason there are now fourteen.

1. Original. One cannot act upon that which has not been defined.

Reconciled. One cannot act together upon that which has not been defined outside of somebody's head. One person, acting alone on a problem they have seen a thousand times, can and should.

A domain narrowing. The original claimed too much, because a definition can live entirely inside one expert and work: recognition has a test, a criterion, and a yes-or-no answer, implemented in tissue rather than in a document. What it lacks is transferability, inspectability, auditability, and replaceability, and those four are what organizations run on and individuals mostly do not need. The rule applies at the moment work crosses from one person to another, or from one moment to a later one where consistency matters.

2. Original. Define every variable to its deepest identifiable condition.

Reconciled. Define every variable to the depth at which further decomposition would no longer change what anyone does with it. That depth is set by the decision, not by the variable, and it moves when the decisions move.

A domain narrowing plus a cost. Deepest identifiable condition names nothing, because atomicity has no absolute meaning: any value decomposes further for some purpose. The floor is requisite variety, which is a real constraint rather than a preference, since a definition with fewer distinctions than the world produces will answer some situations with an action built for a different one. The ceiling is the decision test. Between them is the working band, and it is narrower than I would like.

3. Original. Don't make the specific unnecessarily vague.

Reconciled. Don't make the specific unnecessarily vague, where you know the specific, the reader needs it, and no second reading is wanted. Records of fact, not statements of value.

A domain narrowing, and a substantial concession. Ambiguity in a mission statement, a negotiation, a diplomatic communiqué or a contract clause built to survive the unforeseeable is frequently the competent choice, and the rule has no business there. Inside a record of a particular thing that happened to a particular person on a particular day, none of those functions is present, and the vagueness is not strategic. It is the absence of a decision rather than one.

The practical form is not an instruction to writers. Specificity is mostly a property of what people are asked, so the lever is the question.

4. Original. The more we standardize the variables and reduce ambiguity, the less complicated things become.

Reconciled. Where cause and effect are knowable, standardizing the variables and reducing ambiguity shrinks the number of situations anyone has to understand, and makes deviation visible. Where cause and effect are clear only in retrospect, it suppresses the variation that is the only instrument you have. Every standard needs a path for the case that does not fit, and the rate at which that path is used is the measurement that matters.

Domain, cost, and the largest single correction in Part I. The original claimed a monotonic relationship. The reconciled version is domain-dependent, and it adds the exception path, without which the rule builds a system that camouflages its own hardest cases. Standardization does not reduce complexity. It relocates it, out of the process and into the exceptions, which are then handled by people trained not to exercise judgment.

5. Original. Do the same thing the same way every time and you'll get a consistent result.

Reconciled. Do the same thing the same way every time and you will get a predictable distribution, not an identical result. The spread that remains is the process, and it is a measurement rather than a failure. Change the way deliberately, as a designed act, and never let it change any other way.

The only ism whose central promise was simply wrong. Holding the method does not produce sameness; it produces stability, which is more useful and which converts the spread from an embarrassment into a reading. The added clause permits improvement, which the original appeared to forbid, and forbids tampering, which the original permitted by silence. Adjusting a process because of one bad result is the most common way do the same thing the same way every time gets broken by people who believe they are honoring it.

6. Original. Almost everything in life can be reduced to a quantifiable formula. When X happens then the user does Y. If A = B, then C = D.

Reconciled. Almost everything in life can be reduced to a quantifiable formula, where the conditions are observable, the mapping holds still, and nobody profits by manufacturing the trigger.

Three conditions replacing one unexamined word. Almost was carrying every exception I had not enumerated; three of them now have names. The third condition is the one I had never stated and the one that does the most work: the moment a rule governs anything anybody cares about, its conditions stop being observations of the world and become things people produce.

7. Original. If it lives in your head, it should live in the "system."

Reconciled. If it lives in your head, it should leave a trace outside your head: written where the knowledge is explicit and merely unrecorded, recorded as a demonstration where it can be shown but not told, and handed to a successor in person where it can be neither. The system holds the trace, not the knowledge.

The book's title ism, and the one whose premise turned out to be false. The system cannot hold knowledge, because an artifact is inert without an interpreter and the interpretive capacity is the thing that was supposed to move. What survives is a smaller and defensible claim: the trace lowers the cost of the next person rebuilding their own version. Judged against transfer, every knowledge-management effort in history has failed. Judged against rebuild cost, they routinely succeed.

One of the three bands does not resolve to a system at all, and any version of this rule with no room for keep the person will quietly destroy something it cannot see.

8. Original. We shouldn't PULL information, we should PUSH information.

Reconciled. Nobody should have to already know that a fact exists in order to receive it. So push what changes an action, to the one person whose action it changes, on a channel reserved for that kind of thing. Everything else is made findable, not sent. Attention is the budget, and every push spends it whether or not the push was worth making.

The largest revision in the book, and a rewrite rather than a narrowing. The diagnosis was right: requiring somebody to know a fact exists before they can get it concentrates failure on the people least equipped to avoid it. The prescription was wrong, because delivery is not the solution to a discovery problem. A push system past its recipients' capacity delivers exactly what a pull system delivers, which is nothing, while reporting that it is working.

9. Original. An efficient system minimizes the amount of user input while maximizing the amount of automated output.

Reconciled. An efficient system minimizes the amount of user input while maximizing the amount of automated output, subject to two limits: it does not remove the contact a person needs to keep a working model of the situation, and it never presents a guess in the same form as a certainty. Where the system is inferring, it shows what it inferred and from what.

A cost and a person. The original priced the user's effort and nothing else: not their attention, not the system's confidence in things it inferred, not an error with no human standing in front of it. The distinction between input that is friction and input that is contact is the one the original could not make, and removing an input removes both.

10. Original. The more we reduce life's decisions to quantifiable formulas, the more automated we can make the simple tasks — freeing the human mind for true, ambitious deep thinking.

Reconciled. The more we reduce life's decisions to quantifiable formulas, the more automated we can make the simple tasks, and the more capacity is released. Where that capacity goes is a separate act, undertaken deliberately, defended against the many claimants who are better equipped than you to collect it. Released and unallocated, it goes to them. Directing it toward true, ambitious deep thinking is a choice about what a life is for, and I am making that choice rather than reporting a result.

The original joined an empirical claim and a value commitment with a dash, in a way that made the second look like it followed from the first. The revision separates them and keeps both. The release is well supported. The destination is not automatic, and a century of enormous productivity growth arriving without the fifteen-hour week is about as clean a demonstration of that as history offers.

11. Original. Where a data-based decision will not produce a genuinely better result, leave it manual.

Reconciled. Before leaving a decision manual, name what you know that is not in the data, specifically enough to be wrong about. Capture whatever can be captured. Leave manual only what remains, and say which reason applies: the information cannot be captured, or it could be and you value something more than the accuracy. Both are legitimate. They are not the same, and only the first is a claim about the world.

A permission converted into a procedure. The original was unfalsifiable, because genuinely better was assessed by the person who did not want to do the work, and the evidence says that assessment is wrong far more often than it is right. What rescues it is the condition under which people actually beat formulas, which is not better judgment but more information, and which is an inventory question with an answer.

12. Original. Identify your bias, then set it aside.

Reconciled. Identify your bias, then set it aside. Where you cannot set it aside, name its direction, so that the pattern of what you do and do not do is legible to somebody else.

The one rule the book applies to itself, and the only change is an admission. Some biases do not get set aside. Mine is that I find converting a decision into a rule satisfying independently of whether the conversion was worth making, and the satisfaction arrives at the building rather than at the being-right, which are separated by years. That predicts which parts of this creed I will follow and which I will skip, and naming the direction is the only correction available.

13. Before changing how work is done, name what the current arrangement does for the people doing it, and count that as part of the design rather than as resistance to it.

This is the thirteenth ism and it came out of a coal mine in 1951. The original creed had twelve technical rules and no social term, and not because the social term was considered and set aside. It did not occur to me. Every unmeasured loss in Part I was a rediscovery of the same thing one symptom at a time: the picture that stopped being formed, the judgment that stopped being exercised, the capability that decayed, the conversation that stopped happening. A design that optimizes the technical dimension while damaging the social one underperforms a design that does both, on the technical measures.

14. A rule that governs a person should be readable by that person and reversible by them. Otherwise it is not a formula they installed. It is a condition they live in.

The fourteenth, and it exists because the method in this book is indifferent to whose interest it serves. The largest, best-funded implementations of exactly this creed are aimed at converting people's deliberate attention into automatic responses for somebody else's benefit, and nothing in the first twelve rules ruled that out. This one does, at least for what I build: it is a constraint stated from the chair of the person decided about, which is a chair the creed never occupied.

Both of the new rules have a property the original twelve do not, and it is worth naming because it changes how they should be used. The first twelve are instructions about how to build. These two are conditions on what you are allowed to build, which means they cannot be traded off against efficiency the way the others can. A definition can be shallower than ideal because the deeper version costs too much; that is a legitimate engineering judgment. A rule the people governed by it cannot read is not a cheaper version of a readable rule. It is a different object.

I am aware that stating it that way makes it sound like a principle I have always held, and I have not. I have built rules people could not read, for reasons that seemed sufficient at the time, mostly that explaining them would have invited arguments I did not want to have. Those arguments were the mechanism by which I would have found out the rules were wrong.

What the fourteen amount to, read together, is different from what the twelve amounted to.

The twelve described a conversion, and they were confident about it, and every one of them was pointed at the same outcome: fewer decisions, held more consistently, by fewer people, taking less time.

The fourteen describe the same conversion with a bill attached. Nine of them now carry a condition about where the rule holds. Six carry an ongoing cost. Three carry an obligation to somebody the rule acts on. And all of that converges on a single sentence that appears nowhere in the original creed and is the most heavily supported claim in the book:

The ratio is excellent. Twenty checks a month against a rule running thousands of times is nothing. What it is not is zero, and the entire difference between a creed that works and a creed that fails in year three is whether that line item gets funded or gets described as overhead.

An organization that takes the freeing and skips the reclaiming has implemented half of this book, which is the half that appears in the business case.

A note on how to read the two versions together, because I do not want the originals treated as drafts.

The short forms are still the ones I say. If it lives in your head, it should live in the system is eleven words and it starts a conversation. The reconciled version is fifty-three words and it finishes one. Those are different jobs and a creed needs both: something short enough to reach for in the moment, and something exact enough to defend when somebody pushes on it.

What the short form must not do is get used as though the long one did not exist. That is the specific failure this whole book is a correction for, and it has a signature: somebody invokes a four-word rule to end a discussion, and the thing they have ended is the part where the rule's conditions would have been checked. Every one of the fourteen has conditions now. A person quoting the short form is making a claim that those conditions hold, whether or not they know it.

So the honest reading is that the left column is a handle and the right column is the object. Reaching for the handle is fine. Believing the handle is the whole thing is how a creed becomes a set of slogans, which is what mine was for about fifteen years and is the reason this book took the shape it did.

Fourteen rules is too many to hold, and I want to say what I think actually gets carried, because a creed nobody can recall is a document rather than a creed.

In practice I do not run fourteen rules. I run three questions and a habit, and the fourteen are what those three questions expand into when somebody asks me why.

The first question is what is this made of. That is Family A, collapsed. What are the terms, do they mean one thing, are they deep enough to tell apart the cases that need different answers, and would somebody else reading this arrive where I did. Almost every argument I have ever watched fail to resolve was failing here, and the tell is always the same: the restatement that does not help.

The second question is what happens when this runs a thousand times. That is Families B and C together, which is not how I have them filed and is how they behave. A thing done once is not a system. The interesting properties all appear at volume: the spread becomes visible, the drift begins, the knowledge concentrates in whoever does it most, the notifications accumulate into noise, the derived field starts being accepted without reading. None of those is detectable in the first instance and all of them are inevitable by the thousandth.

The third question is who is this for, and what does it cost them. That is Family D plus the two new rules, and it is the question I did not used to ask at all. Not whether people will like it. What the current arrangement is doing for them that the new one will not, whether they can read the rule that governs them, whether they can turn it off, and where the time it saves is going to go.

And the habit is the one the method rule names: assume I am biased toward building, because I am, and correct in the direction of not building.

Those three questions are in a deliberate order. The first is cheap and settles most disputes. The second is where the money is, and it cannot be answered in the first instance, which is why it gets skipped. The third is the one that never gets asked at all, and it is the only one that reaches the people the work is done by rather than the work itself.

Running them in the other direction is how good designers build things nobody wanted. Start with who it is for, and you get a rule that is kind and does not survive volume. Start with volume, and you get a rule that scales and nobody will use. Start with the terms, and at least you find out early whether there is anything here to build.

That is what fits in a head. The fourteen are the audit trail.

One more piece of accounting, because the pattern of what changed says more than any individual change.

Eleven of twelve were narrowed. Nine gained a domain condition, which is to say a statement of where the rule stops being true. Six gained an ongoing cost. Three gained an obligation toward the person the rule acts on. One was rewritten outright rather than narrowed, and one had its central promise withdrawn and replaced with a smaller true one.

What did not happen, and I half expected it to: none of the twelve turned out to be empty. Every one of them was pointing at something real, and every one of them was stated in a form that claimed more than the something. That is a consistent enough pattern to be worth a sentence of its own. The isms were not wrong. They were unconditioned, which is what a rule looks like when it was built from experience and never tested against the cases it did not happen to meet.

That also predicts which ones a reader should distrust most, and the prediction is not the one I would have made at the start. The rules I was most confident about are the ones that took the heaviest damage, because confidence is what stops you looking for the boundary. The two I was least sure of, the ones about purpose and about what to leave alone, came through better than the ones about definition and standardization, because uncertainty had already done some of the narrowing before the research arrived.

One thing this list cannot show, and it is the most useful thing I learned making it.

Every reconciled version above was produced the same way: by finding the case the original did not cover, and then discovering that the case was not exotic. Not a thought experiment, not a corner of the world I would never touch. A drive-thru on Route 8. A weekend list. A dispatcher who knew four facts. A form field that arrived pre-filled with the wrong building. Every single narrowing in this chapter came from an ordinary case sitting in plain sight, and the reason none of them had narrowed the rule already is that a rule you believe does not get pointed at the cases it would fail.

That is a claim about how creeds go wrong and it generalizes past mine. A rule built from experience is built from the experience you had, and the experience you had is filtered by what you were already doing, which was filtered by the rule. The loop closes and the rule stops being tested by anything, and it can run for fifteen years in that condition looking exactly like wisdom.

The only thing that broke the loop here was going and reading what people who disagreed had actually found, which is a slow and slightly humiliating process and is the only method I know. Nothing in the fourteen rules would have produced these fourteen rules. The creed cannot revise itself; it took an outside standard, and the outside standard was other people's evidence.

Three things did not get resolved and are not going to be resolved in these pages.

Deep thinking has no definition, which means the one ism that says what all the others are for is the one that cannot be evaluated by the standard the rest of the book applies. I could manufacture a proxy. Almost everything in life can be reduced to a quantifiable formula turned into a long argument about what happens to proxies that govern anything anybody cares about.

The definitional regress has no bottom, and the creed stops it by convention: a term is defined when a test, a criterion, and a yes-or-no decision exist. That is a stopping rule adopted for convenience, and every definitional system rests on terms somebody agreed to quit questioning.

And a rule can pass every test in this book while being pointed at the wrong quantity. Defined deeply, held consistently, maintained faithfully, readable and reversible and jointly optimized, and wrong about what matters. Nothing here catches that, because everything here compares a thing to its own specification.

That last one is the reason this is a creed rather than a method. A method can tell you how to build the thing right. What the thing should be is a question you answer with the part of yourself the machinery was supposed to free up, which is the whole argument of freeing the human mind for true, ambitious deep thinking, arriving at the end where it belongs and still without a test.