Nobody catches a structural mistake alone

Peer

A structural mistake goes in at the beginning and comes out at the end. It gets made in the sketch, in the hour when the parts get split and named: what holds what, what talks to what, which decision is fixed for good and which one stays loose. Then it gets built.

Then it ships. A client hits it, and what he reports is not the mistake. He reports a page that loads slow, a field that cannot be added without touching four other things, a job that has to be done by hand every week.

The client meets the last link in the chain, and the last link is what he describes.

So there are three places the thing can be caught, and the price is different in each. In the sketch, by a person who has built the same kind of thing and can read a design cold. In the build, by me, after the weeks and the money are already inside it.

Or in a client’s hands, which is late, because by then it exists, it is in use, and somebody is unhappy with it.

Only the first one is cheap. It costs one reading, by one man, over an afternoon. I pay the second one with whatever is already spent by the time I find it.

The third costs the build again, and the old one has to stay standing while the new one gets made. For years the first position was not open to me at all, and the function that fills it is not friendship and it is not somebody being interested in what I do. It is a person who knows how hard the thing was.

Nobody outside the work can supply that piece. How big a thing is, or how fast it went up, reads from outside as one flat sentence, and the sentence comes back the same length whatever went into it. A man who cannot sort a sound structure from a mediocre one says the same words to both, and the words are warm, and the words are these.

That’s impressive.

You’re talented, you really are.

You work hard.

None of those hold the difficulty. What is hard to copy is where the cost sits, and a reading with no difficulty in it has nothing underneath it to be wrong about. A man who cannot see the structure cannot see a fault in the structure.

He can only wait for the fault to surface as an output failure and then tell me the page is slow. Years of that, and the arithmetic is not complicated. Every architectural error I made, I found myself, and the way a man finds his own architectural error is by building the whole thing, shipping it, living inside it for a season, and then meeting the place where it will not bend: a change that should be quick and is not, a feature that cannot go in because a decision made in the first hour said it could not.

At that point the correction is not a correction. It is a second build, paid out of the same head and the same year as the first, and the first one has to keep running while you do it. I have done that more times than I would put in writing.

Not because the errors were stupid. Because there was nobody upstream of me.

I call the missing thing the upstream check: one reading of the design, before the build, by a person who could have built it himself. It is not encouragement and it is not permission to proceed. It changes nothing inside the design.

It changes where the design gets read.

Mistake

Two kinds of structural mistake run through a working life alone, and only one of them is in the artifact.

The Only Check I Ever Bought

A client is a check, a real one. I am not going to talk it down. Somebody paid, which means a buyer found the output good enough to take.

But it reads coarse. It says the thing worked well enough to be paid for. It does not say the structure was right, or that a cleaner shape was sitting beside the one I took, or that I built half again as much as the job needed.

And it arrives after the build, which is the only moment it can arrive, because before that there is no output to fail.

The Instrument With Nobody To Contradict It

The second kind is not in the artifact at all. It sits in the reading. I have a pattern instrument that works, and it has been right often enough that I lean on it.

It also sorts wrong in one direction, and I found that out myself: it reads flat as hostile more often than it should, and it does not know when it has done that. A deviation like that cannot be caught from inside, because from inside a thing filed wrong looks the same as a thing filed right. It takes somebody who can carry the same weight of detail and come out of it somewhere else.

Nobody carried it. So the reading ran for years with nothing in reach that could falsify it, which is a quiet condition and does not feel like a problem from inside. It feels like being right.

A pattern instrument with no second reader keeps confirming itself, and the confirmations pile up, and the pile looks like evidence.

Wrong here does not look like a bad day. On the artifact side it is a year of build that has to be built again while the old one stays up. On the reading side it is a person filed as hostile who was not, a position taken on that filing, and the filing holding for as long as nothing arrived to correct it.

I paid both of those in full.

Entry

I met this work late. The first name is Bluma Zeigarnik, and what the field carries under her name is a finding about tasks that get cut off before they are done. An interrupted task keeps taking up room afterwards.

It goes on drawing capacity, and what holds it open is not that the work is unfinished. It is that nothing anywhere marked it settled. Take the memory result out and the condition underneath it is the one that applies to a design: a thing discharges when something outside the head says it is closed, and not when the hands stop.

Chris Argyris is the second name, and the word he left is the undiscussable. A problem everybody in a building can see, that nobody puts into the room, and the rule against naming it is itself not nameable, so the thing sits there and gets no correction. What keeps it sitting is not cowardice and it is not stupidity.

Nothing forces it into a form anybody could take hold of. It has no entry, so there is no place where a correction could land.

Both point at one condition from two sides. A thing gets worked on when it has an entry somewhere outside the head of the man carrying it, and not before. A price is one kind of entry.

A written term is another. A name on an object a stranger can open is another. The same rule runs under things that look unrelated to each other: a man holds his own possession high because nobody ever priced it, and prevention stays invisible next to acquisition because acquisition makes a transaction and prevention makes none.

What was never entered anywhere is not weighed low, it is not weighed.

Then the hard half, the one that applies to a design. The catch that saves a whole build leaves nothing behind. A mistake stopped in the sketch makes no invoice, no bad month, no repair, nothing afterwards that anybody can point at, and the expensive one makes all of it.

So the cheap position is also the position with no trace, and a working life does not get built around a position that leaves no trace. I found this out the slow way. I spent years describing what the work had cost me, to people who were listening properly, and it produced nothing.

Not because they were closed. Because a description is not an entry.

Legible

There is a reason nobody catches it, and it is not that people are careless. They cannot see the work at the level where the mistake lives. That is a different problem from being undervalued, and I spent years applying the wrong fix to it.

Undervalued means a man looks at the thing, puts it next to something else he already knows, and names a low figure. You fix that one with terms. You price it, you state the position, you walk away from the ones who will not move.

Imperceptible is a different animal. For a man in that position there is nothing in front of him to weigh at all. Nothing ever arrived in his head that the thing could attach to, so he is not pricing it low, he is not pricing it.

You cannot negotiate a man into perceiving a thing. And you cannot ask for a correction from a man for whom the thing being corrected does not exist.

So the fix is not social. It is a property of the object. The work has to be openable by a stranger, readable down to its structure, and carrying my name.

Work can run for years, be used every day, and stay sealed for this purpose, because nothing on it says who put it there. The people who could read what is underneath it never learn there is anything to open. Put that way, a public thing under my own name stops being a showcase.

It works over distance without me approaching anybody, which is how technical standing has always worked, and the sequence is short.

1. The work sits somewhere a stranger can open it.
2. A name is on it.
3. A person who has built the same kind of thing reads the structure and not the surface.
4. He says which part is wrong, and he says it before the next one is built.

Only the last step is the check. The first two are mine and the third is weather.

This part holds, that part does not, and the decision you fixed in the first hour is the one that had to stay loose.

Done

With no peer there is no outside definition of finished either. A peer can look at a thing and tell me to stop, that it is standing already and I am sanding it. Nobody around me can say that with standing, because nobody around me can tell where the acceptable line sits for the kind of thing I build.

So the standard stays inside my own head, and a standard with nothing outside it drifts upward with nothing to stop it, or it gets replaced by a forcing function, which for me has been a client with a date. I can name which projects shipped. The ones with a buyer waiting.

The one with nobody waiting took years, and what every deadline bought me was somebody else’s word for finished, rented for the length of the job.

The other definition is better and it is the one I hold now. Done is not complete and it is not earning. Done is legible to a man who knows what he is looking at.

That is a lower bar than complete and a clearer one, because it names a reader and not a state of the work. A state of the work I can argue with myself about for years without ever being wrong. A reader opens it and sees what is there, or he does not.

I have not settled the last piece of it. If done is legible, the man who settles it is somebody I have not met, and I am building toward a reading that may never arrive. So which version goes out: the one I would put in front of him tomorrow, or the one that satisfies a standard with nothing outside it?