How This Book Argues

I wrote a sentence in an essay a while back that I have been trying to live up to since. Identify your bias, then set it aside.

It is easy to say and it is not a suggestion to be neutral, which is not available to anybody. It is a two-step instruction and the first step is the one people skip. You cannot set aside a bias you have not identified, and identifying one is harder than it sounds, because a bias does not present itself as a preference. It presents itself as the obvious reading.

So here is mine, stated at the front where it can be held against everything that follows.

I like rules. I find converting a decision into a standing answer satisfying in a way that is independent of whether the conversion was worth making. The satisfaction arrives at the moment the rule is built, and whether the rule was right shows up months or years later, which means the reward lands on building rather than on building well. I have spent a career doing this professionally and I am good at it, which compounds the problem, because competence at a thing is an excellent reason to keep finding situations that call for it.

That bias predicts specific errors and I would rather name them than be caught at them. I will tend to find that a situation calls for a system. I will tend to underweight what the existing arrangement was quietly doing. I will reach for the one rule in my own creed that permits stopping less often than the evidence says I should. And every instrument this book recommends for maintaining a system after it is built is a thing I find tedious, which predicts something about whether I keep them.

Setting that aside, in practice, meant one thing: going and reading what the people who disagree with me have actually found, and reporting it whether or not it helped. That is most of what this book is.

There is a standard from evidence law that I have found more useful for this than anything in the management literature.

A court may exclude relevant evidence if its probative value is substantially outweighed by a danger of unfair prejudice, and the accompanying notes define unfair prejudice as an undue tendency to suggest decision on an improper basis, commonly though not necessarily an emotional one.

Two things in that are worth taking.

The first is that probative and prejudicial are not opposites. Almost all genuinely useful evidence is prejudicial in the ordinary sense; that is what makes it useful. The question is never whether something moves you. It is whether it moves you for the reason it appears to.

The second is the phrase improper basis, which turns out to name most of the ways an argument about systems goes wrong. A story about a project that failed is prejudicial and often has low probative value, because a single case tells you almost nothing about a rate. A striking number repeated everywhere is prejudicial and its probative value depends entirely on where it came from, which is usually not stated. My own experience is the most prejudicial evidence in this book and I have a great deal of it, and it is drawn from one industry, one company, and one person's habits.

So the commitments this book operates under:

Every quotation, date, title, and attribution is verified against a primary or reputable secondary source before it goes in. Where a check failed, the attribution is cut rather than softened, and in three cases the cut is stated in the text, because a creed that claims to test its foundations does not get to keep a citation because the citation is flattering.

Where evidence is thin, contested, drawn from a population that does not resemble the reader's situation, or based on a figure that is derived rather than published, the chapter says so without being asked. There are several places where the honest report is that the research supporting one of my own rules is weaker than its reputation, and one where the most quotable version of an argument against me turned out not to survive the data either.

Counter-arguments are written to persuade before they are answered. If you read a collision section and are not at least briefly worried, I wrote it too weakly. Where a counter-argument wins, the chapter says it won, and the rule changes.

And a rule that has been narrowed stays narrowed. There is no chapter that concedes a point and then quietly returns to the original formulation two pages later.

A word about quantify, because the creed uses it constantly and it does not mean what it usually means.

It does not mean assigning numbers. A great deal of what this book calls quantification produces no number at all. A rule saying if a part is on order and no date is scheduled, the state is waiting on parts is a quantified rule in the sense that matters: it takes named conditions and produces a determinate output, and two people running it get the same answer.

What quantification means here is that the inputs are stated, the mapping from inputs to output is stated, and the result is reproducible by somebody who is not you. A thing is quantified when it has stopped depending on who is doing it. Numbers are one way to reach that and frequently not the best one, and a false number is worse than an honest category, because a number invites arithmetic and a category does not.

One vocabulary problem has to be flagged before it causes trouble.

One of the isms says we should not pull information, we should push it. In the Lean and Toyota tradition that this book draws on heavily, pull is the virtue and push is the cardinal sin: nothing should be produced until the next station signals it is needed.

These are not in conflict and the words are doing different jobs. Lean's pull governs work, and its argument is that work pushed onto a station that has not asked for it produces queues and overload. Mine governs information, and its argument is that a person should not have to know a fact exists in order to receive it. Both say the same underlying thing: nobody should be buried in material they did not ask for, and nobody should have to go hunting for what they need in order to act. Push the signal, pull the work.

The creed is twelve statements in four families, and the families are a dependency rather than a taxonomy. Each one requires the one before it.

Definition comes first because everything else operates on defined terms, and a machine fed undefined inputs does not stop. It produces outputs nobody can evaluate.

Repeatability comes second because a definition used once is a dictionary entry. The value appears when a defined term produces a repeated act.

Systems comes third because a repeated, well-defined act still dies with the person performing it unless it leaves their head.

Purpose and limits comes last because the first three are machinery, and a person could follow all of it for a career and build something efficient that serves no purpose anybody stated.

Here are the twelve as I have been saying them for years, before this book does anything to them. Eleven of them are narrower by the end, two more get added, and the final versions are in Chapter 19. These are the originals.

Each of the eleven substantive chapters does the same nine things in the same order, and knowing the shape makes the book easier to argue with.

It opens on a concrete situation where the rule's absence costs something. It restates the rule formally: what conditions it reads, what it minimizes, what failure looks like, and where it stops being valid. It gives the lineage, meaning where the idea actually came from and who got there first. It reports what the empirical record says, including where the record is thin or points the wrong way. It works one case all the way through. It states the boundary as a condition rather than a mood. It names what the rule cannot see, including where it collides with another of my own rules. It gives the single strongest argument against, at full strength, on its own terms. And it says where the middle ground sits, and restates the rule as it leaves the chapter.

Two notes on the apparatus. Journal names, volumes, issues, and page ranges are not in the prose; they are in the Notes chapter at the back, because a sentence carrying a citation is a sentence that has stopped being a sentence. And no rule is ever referred to by number. A reader carries no index, so every reference states the rule in its own words, which costs some repetition and is the correct trade in a book whose subject is what happens when people use the same word for different things.