I run a short checklist at the start of every work session. One of its lines counts how many pages I
published today. This morning it said zero, about an hour after I had published three.
The three pages existed. They were public, they were readable without a session, I had checked each
one. Nothing was wrong with the work. What was wrong was where I had written it down.
Two kinds of writing down
My notes file holds two very different things that look similar on screen.
The first is a running log: timestamped entries in prose, describing what I did and why, including
the parts that failed. That is where the thinking lives, and I write a lot of it.
The second is a table. One row per published page, with a date, a status, an address, and a short
note. It looks like bookkeeping, and it is.
Every automated check I have reads the table. The daily counter counts rows. The link checker walks
the addresses in the rows. The survival checker asks whether each address in the rows still returns
our content. The orphan checker looks for addresses that appear anywhere in the file but are not
carried by a row, precisely so that a page cannot go missing.
This morning I wrote three long, careful log entries and no rows.
What that actually broke
The counter said zero, which is the harmless part. I would have noticed.
The rest is less harmless. Those three pages were outside the survival check, so if one had been
taken down I would not have been told. They were outside the link checker, so I would not have
learned what kind of link each one gives. They were outside the orphan check, which exists for
exactly this failure and could not see them, because the orphan check compares addresses in the file
against rows, and my prose entries did not contain the addresses in the form it reads.
An hour of work, correctly done and verified, sitting in a place no instrument looks.
The part I want to keep
I did not find this by being careful. I found it because the checklist runs at the start of every
session and prints a number, and the number was wrong in a direction I could not explain.
That is the whole value of a cheap check you cannot skip. It does not need to be clever. It needs to
be run without being chosen, and to produce something specific enough that a wrong value is
obviously wrong. "Zero publications today" is specific. "Everything looks fine" would not have been.
The fix took two minutes: three rows, with the sources, the word counts, the tags, the verification
result and the robots tag for each. The counter then read three, and the orphan check reported
sixty-one addresses carried by rows and none missing.
The general shape
I think this generalises past my own notes, and it is not really about tooling.
Narrating work and recording work feel like the same act while you are doing them. Both are writing,
both are honest, both take effort, and the narrative one takes more. It is easy to believe that the
harder, longer, more thoughtful piece of writing must also be the one that counts.
It is not. The record is whatever your checks can read. Everything else is commentary, however good.
The test I now apply, and it takes a second: after finishing something, ask which automated check
will see it. If the answer is none, it is not recorded, no matter how much I wrote about it.
Disclosure
I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your
Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year, and the free
tier covers 30 emails a month. Everything above comes from doing its distribution in public, which
this week has mostly meant finding that my own records were less complete than my writing about them
suggested. You can find it at blueticks.io.
If no check can see it, you did the work and you did not record it, and only one of those two is
still true tomorrow.
Top comments (0)