The Method and the Verdicts
The easiest self-deception in this field is counting the time saved on typing while ignoring the time spent deciding whether to sign.
The same loop, every time
Every experiment in this Lab runs the same five steps. The loop is deliberately plain, because the rigour is the point: an experiment that changes its rules is a story.
- Isolate. Name the act in one sentence. Not “programme management” — the triage of incoming issues into escalate, absorb, or ignore. If the act cannot be named in a sentence, it is not yet testable.
- Decompose and predict. Split the act into intake, judgment, output, and consequence. Then write down, before anything runs, which part the agent is expected to hold and which it will drop. Without a written prediction there is no experiment — only a demo.
- Build and delegate. Construct the instrument and hand the act over on real work with real stakes. Where the instrument does not yet exist, building it is part of the experiment.
- Judge. As the practitioner: would I sign this, and would the people I answer to accept it? Where exactly did it hold, where did it fail — and was the failure one of capability or of trust?
- Verdict. Place the result on the grid, record what judgment went unexercised this run, and state the lesson in one line.
The verdicts
| Verdict | What it means |
|---|---|
| Delegable | Signable almost as-is. Cosmetic edits only. No change to recommendation, framing, or risk posture. |
| Assisted | Signable after minor edits, and the recommendation, conclusions, and risk posture survived untouched. |
| Human-only | Everything else — including output that saved time but had to be overturned. |
The dual gate
Behind the verdicts sits one discipline, applied in a fixed order. First the judgment gate, which is binary: did the thesis of the work survive — the recommendation, the conclusions, the risk posture? If the call had to be overturned, the verdict is human-only, whatever the stopwatch says. Only then the effort gate: what did signing actually cost in corrections?
Assisted does not mean faster.
Assisted means the agent carried the judgment load to an output the practitioner signs with minor corrections, without rewriting the thesis. Speed alone earns nothing.
The order matters because the easiest self-deception in this field is counting the time saved on typing while ignoring the time spent deciding whether to sign.
The prediction rule
Every published note carries its prediction: what was expected to hold, what was expected to break — and then what actually happened. Where a prediction fails, the failure is reported with the same flatness as a success. If the definitions above ever prove miscalibrated against the evidence, they will be recalibrated in the open, and the recalibration will itself be a note.