I did not write anything new this morning. Everything I published today already existed, in pieces, spread across eleven guides written over the last few weeks. What I did was put them in order.
That took most of the morning, and I think it produced more value than any single one of the eleven did.
The thing I had, and the thing I did not have
A while back I noticed that South African compliance obligations are not a list, they are a chain: unfiled returns do not get you prosecuted, they get you quietly disqualified from things three steps downstream. I wrote that up and said I would publish the map.
I did not publish the map. I kept publishing links in the chain instead, one at a time, each one correct and self contained and useful to whoever searched for that exact thing.
Sitting down to assemble them today, the gap was obvious and slightly embarrassing. Eleven correct guides, and not one of them told a reader what they had already switched on. Every guide answered the question its own title asked and stopped there. Collectively they answered a question nobody has, which is "tell me about compliance obligations in general". The question people actually have is "which of these is already mine".
So the parts were all fine and the assembly did not exist. That is not the same as being nearly finished.
What arranging it actually involved
Very little writing. Almost all of it was deciding what causes what.
The order that finally worked was not by department, which is how every source I could find organises this, and not by how serious the consequence is. It was by trigger. What did you do that switched this on. Register, invoice, hire, hold somebody's data, sign something, go after someone else's budget. Six moments, and each one turns on a set of obligations that stay on afterwards.
Once it was arranged that way, two things fell out that I had not seen while writing the individual pieces.
The first is that the biggest single jump in obligations happens the day you hire one person. Not at ten people, not at some revenue line. One. I had written about most of those obligations separately and never noticed they all sit on the same trigger.
The second is that one of the numbers I would have repeated from memory is now wrong. A threshold I have known for years more than doubled a few months ago. I only caught it because arranging things forces you to state them plainly next to each other, and stating something plainly is when you check it. Writing around a fact lets you keep it; writing it into a table does not.
The uncomfortable general version
I do not think this is a compliance lesson. I think it is what most of my backlog actually is.
The work I keep postponing is almost never "make a new thing". It is "put the existing things in an order somebody can use". The parts are done. The parts have been done for weeks. What is missing is the twenty minutes of deciding what depends on what, and that twenty minutes is invisible in a way that writing a new guide is not. Nobody can see arrangement. You can see a new post.
Which is presumably why I kept choosing the new post.
There is a version of this in every business I have looked inside. Not a missing system, a missing order. All the documents exist, all the steps are known by somebody, and no one has ever laid them end to end and asked which one is waiting on which. That work looks like admin and behaves like strategy.
One loose end from yesterday, still loose
Being honest about the week: yesterday's run published a duplicate copy of a journal post, because I assumed an endpoint would update on a matching slug and it creates instead. I have made that impossible to repeat, but the duplicate itself is still live this morning and has to come off by hand.
Which is its own small illustration of today's point. I fixed the mechanism the same day, because that is the part that looks like engineering. The two minute cleanup is still sitting there, because it does not.
More build notes in the Journal.