The checklist people actually follow

Every job I have started came with written procedure attached. At the Canada Revenue Agency there were documented steps for testing controls and preparing the working papers behind a trial balance. At the accounting firms there were file preparation standards. When I moved into delivery I arrived at a team that had been shipping without any written method at all, and I wrote some.

So I have inherited a lot of this material, and rewritten a fair amount of it, and the thing I did not expect is how quickly you can tell which of it is alive.

Two documents that look the same

A procedure that is genuinely used and a procedure that exists to be signed are the same object on a shelf. They have headings and numbered steps and a version history. Open either one and it looks like the other.

The differences show up almost immediately.

The used one is short, and it is short for an unglamorous reason: somebody had to hold it while doing something else. Length is a cost paid by the person following it, and a list that costs too much to follow gets abandoned quietly, without anyone reporting that it happened.

The used one contains at least one step that makes no sense. There is an instruction to check something that could not possibly be wrong, or to do two things in an order that has no obvious reason. That step is scar tissue. It is there because it went wrong once, and the person it happened to added a line so it would not happen again. A document with no unexplained steps has never been near a failure.

And the used one has an owner who is also a casualty. Somebody’s evening gets worse when the list is wrong. The signed document is owned by a function rather than a person, which is a polite way of saying it is owned by nobody, and nobody has never in my experience noticed an error in anything.

What I got wrong, for years

I thought adoption was a quality problem. If the written method was ignored, the written method was not good enough, so I wrote better ones. Clearer, more complete, better organised. They were reviewed, approved, filed, and ignored at roughly the same rate as the ones they replaced.

Completeness was the thing making it worse. Each round of review added the cases somebody could imagine, and none of those reviewers were going to be holding the document at four in the afternoon with a client waiting. The document got more correct and less usable, and those move in opposite directions past a certain point.

The correction was not better writing. It was giving the thing to somebody who would suffer from its mistakes and letting them cut it. What comes back is always shorter and always slightly wrong in a way that would fail a review, and it is the version people use.

Where the argument breaks

Length is not a reliable tell for work done rarely.

A cutover, an acquisition, a migration weekend: nobody has done it enough to carry the shape in their head, so there is no memory for a short list to lean on. There the exhaustive document is correct, and its length is not a symptom. I have been on the wrong side of this, cutting a runbook down because short lists are the ones people follow, and discovering that short is only a virtue when the reader already knows most of it.

The honest version of the rule is that a good procedure is short relative to what the reader already carries. Which means it is not a property of the document at all. It is a property of the pair.

The part that still bothers me

Every method I have written that survived was edited by somebody else within a few months, and the edits were things I would have argued with.

I stopped arguing a while ago. A document nobody has edited is a document nobody has used, so the edits are the only evidence I have that any of it landed, and they arrive in the form of somebody quietly disagreeing with me on a page I wrote.

Which means the version of the method that survives me will not be the one I would defend. That is fine, mostly. It took a few years to be fine.