The Real Work Instruction Was in a Notebook in His Toolbox

The official work instruction for that cell was eleven pages, laminated, in a binder on a stand beside the machine, revision F, dated 2014. The actual work instruction was a spiral notebook in the operator's toolbox, and it was about forty pages, and roughly a third of it was crossed out.

I only found out about the notebook because he was out for a week with a back and the cell scrapped nineteen parts in two days, and when he came back he looked at the scrap, said one word about the fixture, and it stopped.

I have thought about that notebook for six years and I have come to think it is the single most important object in most manufacturing plants and the one nobody has any theory of.

Start with what was in it. Not secrets, and not tricks. It was a record of every way the official instruction was wrong, accumulated over eleven years. The instruction said torque to a value; the notebook said which of the four fixtures ran high and by how much, because the fixtures were not identical and had never been identical. The instruction said inspect for burr; the notebook said where the burr shows up first when the tool is getting dull, which is a specific corner, and gave you about forty parts of warning. The instruction said nothing at all about the compressor, and the notebook had a line about not running the cell in the first twenty minutes after startup in winter, which turns out to be about air moisture, and which nobody had ever written down anywhere else in the building.

Every one of those entries exists because the official instruction failed and a person noticed and made a note. The notebook is not an alternative to the document. It is the document's maintenance log, kept privately, because there was no public place to keep it.

Now, the case against me, and it is strong enough that I have argued it myself in front of a plant manager.

Undocumented knowledge in one man's toolbox is a liability of the first order. It walks out of the building when he retires, which in Ray's case was going to be about four years away. It cannot be audited, so nobody can verify that the thing he does is safe, and in a regulated industry that alone ends the conversation. It cannot be taught systematically, so the next person spends eleven years rediscovering it, badly, at some cost in scrapped parts and possibly fingers. And it makes the plant dependent on an individual in a way that is bad for the plant and, less obviously, bad for him — Ray could not take a holiday without something going wrong, and he knew it, and that is not a compliment, it is a trap.

Everything on that list is true. Standardisation is not a bureaucratic imposition; it is how an industry stops killing people, and I have no patience for the romance of the irreplaceable craftsman.

So the notebook should be brought into the document. Agreed. That is not what happened, and the difference between what should have happened and what did is the whole of what I want to say.

What happened was a digitisation project. The eleven pages went into a tablet mounted on a stand where the binder used to be. It was a real improvement in several ways I will not dispute — the revision was always current, the pictures were better, and a supervisor could see who had acknowledged the latest change. Three months of work by competent people, and the result was that revision F became a screen.

Nobody asked about the notebook, because nobody outside the cell knew there was one. The project scope was to digitise the work instructions, and the work instructions were the binder.

And here is what the tablet did that the binder had not. The binder was a piece of paper on a stand and a man could write on it. People did — there were pencil notes in the margins of revision D, which is how I know this was going on before Ray. A screen cannot be written on. The tablet was locked, correctly, because an editable work instruction in a regulated plant is an audit finding. So the only remaining place for a correction to go was the notebook, and the notebook was now the sole copy of the only mechanism by which the official instruction had ever improved.

That is the failure, and it is not that the plant digitised. It is that the plant digitised the artefact and not the process that maintained it. Revision F was not a document. It was the current output of a loop: instruction, failure, notice, note, and — occasionally, when somebody escalated — a revision. Cut the loop anywhere and the document stops being current, and it stops silently, because a document that is not being corrected looks exactly like a document that is correct.

The fix is not complicated and it is not technology. It is a channel: a way for the person at the machine to attach a note to a step, visible to the next person on the shift, reviewed by an engineer on a cadence, with the ones that survive review promoted into the instruction. Every part of that exists in every system sold for this purpose. It is almost never turned on, because turning it on requires somebody to own reading them, and reading operator notes is not a job anybody is measured on.

Ray retired in 2022. I asked him, near the end, what he was going to do with the notebook, and he said he had already given the fixture pages to the second-shift guy and that the rest of it would not make sense to anybody. I said he should hand the whole thing over anyway. He said, not unkindly, that he had tried that in about 2016 with a different supervisor, and had got it back with a note asking him to submit changes through the proper channel.

I have been in the plant since. The cell runs. Scrap is up somewhat and there is a project about it, and the project is looking at the tooling, which is a reasonable place to look, and the compressor comes on at six in the morning in February the same as it always did.

A note on the change: the operator is named in the first version of this piece and asked me afterwards to take it out, which was entirely reasonable and which I should have asked about first. The earlier revision is still readable, which is how this site works and which I agree with, and I am not going to pretend the name was never there.

A year on: the plant did enable the operator-note channel, and it ran for about five months, and it was switched off again because nobody was reading the notes. That is not a refutation of what I argued. It is the objection I raised against myself — that turning it on is free and owning it is not — arriving on schedule.