The question
Am I, the person, still intact? The Cognitive question is the first of the nine because it is the one that built the other eight: the thinking that formed my judgement in the first place, and the focus to hold a hard problem in my head for an hour. On this build the question is whether the code works. On the long clock it is whether the part of me that decides is still doing the deciding, or whether it is being handed over a sentence at a time, and never noticed, because each sentence was small.
Self-report is useless here. Ask me on any afternoon whether I am still thinking and I will say yes, and mean it. So I keep the transcripts, and I read them.
What the line says
Of all the text in my agent transcripts each week, the share I typed rather than the agent produced. In October 2025 it was about 55 percent. It fell through the winter, and in June 2026, the heaviest agent month in the archive, it touched 7 percent for a week. It has climbed back and sits near 37 percent this month.
Beside it sits a second line: for each day with any agent activity, how many repositories I touched. Two a day until May. Four since July. The busiest day touched eleven.
Read alone, each line supports a small claim. Read together they say something I did not set out to learn: the summer the agents carried more of the words is the summer I was spread across more systems. That could be delegation working as designed. It could be the start of the gap between systems touched and systems owned, which is the rate at which ownership dilutes by over-distribution. The line cannot say which. A pasted error message counts as my words. A long instruction counts as my words. Neither is reasoning.
The three questions
Every research post runs the same three rows. Which of the nine it names, how good the evidence is, and whether the thing is being measured.
| Row | Status | Evidence |
|---|---|---|
| NamedWhich of the nine questions does this answer? | Cognitive | The first line reads how much of the conversation came from me; the second reads how thin I was spread. Both are Cognitive signals in the model, and both are near neighbours of locked signals the kit will read under their own names. |
| EvidencedHow good is the evidence? | open | The archive shows the movement and not the cause. A fall in own words is consistent with delegating better and with thinking less, and the two look identical on this line. It raises the question. It does not answer it. |
| MeasuredIs this being read on the Data page? | on Data | Both readings publish with version-zero definitions, one weekly and one monthly, refreshed by script and never by hand. The definition sits beside the chart. |
What this cannot say:
- Whether my words were reasoning or instruction. A typed thought and a typed command count the same. The difference between them is the whole difference between delegating and declining.
- Whether a touch was a two-minute check or a four-hour build. Eleven repositories in a day might be eleven systems I own or ten I glanced at.
- Whether the recovery is me changing or the work changing. The summer's projects were different from the spring's. The line does not know that.
- Anything about what shipped. No reading in the archive links an exchange to a commit or a deploy. This is a line about the conversation, not the output.
What I am doing about it
The status says what is missing. These are the moves, on my own work, that would supply it. Each is either read automatically off the footprint or asked of me on a cadence, and each costs something to keep doing.
- AutoKeep the two lines side by side and refuse to read either alone. Own words on its own is a story about delegation. Projects per day on its own is a story about being busy. Together they are a question about ownership, and the page should keep asking it rather than pick the flattering half.Costs: a chart that says question where a reader wants an answer.
- Asked · MonthlyOf the systems I touched in the last fourteen days, which do I actively own? The model asks it monthly for a reason: the honest number is smaller than the touched number, and the gap is the reading. The answer goes on the record and joins the archive like everything else.Costs: an honest list, once a month, in writing, of systems I have stopped owning.
- AutoWhen the kit locks, split my turns into reasoning and instruction. Publish the reasoning share under its own name, not this one. The own-words line stays as the near neighbour it is.Costs: a classifier that will be wrong some of the time, with a definition that says exactly how it decides.
- Asked · QuarterlyA fixed class of problems, solved without a tool, timed. The maintenance signal the model puts first. It is the only one of these that can tell me whether the capacity is still there when the agent is not.Costs: a quarterly afternoon proving something to nobody but me, and the possibility of finding out.
What I keep
The delegation. Four repositories a day is what it looks like when agents carry the middle of the work, and I would not go back to one. This line is not an argument against the practice. It is the instrument that lets me run the practice without trusting my own afternoon read of it.
What it costs is the reading itself: fifteen minutes a week looking at a chart that might bend the wrong way, and a published definition I cannot quietly change when it does.
From my own transcripts, October 2025 to September 2026, archive readings v0. Written with AI assistance; the reading and the judgement are mine. — Clay