Defining quality before anyone asks
The checks were the easy part. The document's real content was one number, and I had no authority to choose it.
I joined a consultancy that had been shipping without a quality method and was asked to write one. Not to improve one. To produce the first.
I want to describe what that is actually like, because I had built plenty of checklists before and I thought this was a bigger version of that, and it was not the same activity at all.
The checks are not the document
Designing checks is craft work and it is pleasant. You go and look at what tends to go wrong, you write down the things that would have caught it, you order them so the cheap ones run before the expensive ones. A morning’s thinking gets you most of the way, and the result looks like the deliverable.
It is not the deliverable. Every check is a claim on somebody’s hours, and the total of those claims is a line: this much verification, no less, before anything leaves. That line is the document. Everything else is the wrapper around it.
And nobody can tell you where the line goes. There is no external body, no number to look up, nothing you can point at that says a piece of work of this kind gets this much scrutiny. There is only a judgement about how much of other people’s time you are entitled to spend to reduce a risk you cannot size, against a schedule that somebody has already promised to a client.
I wrote the first version and realised that the actual sentence I had written, underneath all the procedure, was: I think this is enough and I think that is not. In my own voice. About other people’s work.
The pressure only ever runs one way
The thing I was unprepared for is that the line gets tested in one direction only.
Nobody has ever come to me and asked whether we should check more. Not once. Every conversation about the standard, for as long as I have had one, has been about whether this particular item can be skipped this particular time, and each of those requests is individually reasonable. The date is real. The change is small. The person asking is not being careless, they are being sensible about one case.
So the floor does not get argued down in a debate. It gets argued down in increments, each of which is defensible, and there is no meeting at which it moves. Which means holding it is not a policy activity. It is a weekly activity, and it is conducted by a person, and everyone knows which person.
That is what I mean by calling a quality standard a political document. Not that it is factional. That its content is an allocation, its authority is personal, and it is renegotiated continuously by people who have every right to ask.
Borrowed authority, and the wrong distribution
Two places.
First, I tried to make the line look objective. I described practice, I appealed to what is normal, I used the passive voice everywhere so that the standard appeared to have arrived from somewhere rather than from me. That fooled nobody for a single week. Everybody knew who wrote it. All the borrowed authority did was make it harder to change, because a document that claims to come from nowhere cannot be argued with, so people stopped arguing and started routing around instead. Visible authorship would have been better. An adjustable line that people push against is in better condition than a fixed one they ignore.
Second, and worse, I set it in the wrong places. Not too high overall. Wrong in distribution. I was strict about things that were easy to check and had rarely caused harm, because those were the checks that were easy to write, and I was loose about a couple of things that turned out to matter, because specifying them properly was hard and I told myself the team’s judgement would cover it.
I did not learn that from the document or from any review of it. I learned it from what went wrong, months later, which is the ordinary way and not a flattering one for somebody whose job was to have thought about it in advance.
The revision is better. I do not think it is right, and I have stopped expecting to be able to tell from the inside. The only real feedback a quality floor produces is failures, which arrive slowly, and successes, which are indistinguishable from having been lucky.