Working Procedure

The procedure

1. Step 1

   State the position before the exchange

   Write down what you currently think and why, before the first prompt. Without it there is no baseline against which drift can be detected, and drift is the failure this whole procedure exists to catch.

2. Step 2

   Record the exchange

   Keep the transcript, not a summary of it. A summary written afterwards is already an interpretation, and it is written by the party with an interest in the outcome.

3. Step 3

   Mark the turns

   Where a machine turn carries the argument rather than being absorbed into it, set it apart and attribute it. Unmarked prose is the author's. That single rule is what makes the rest of the record readable years later.

4. Step 4

   Check outside the exchange

   Any claim that will be relied upon gets verified against a source that was not part of the conversation. Internal coherence is not evidence; it is the thing a fluent system produces most easily.

5. Step 5

   Label the standing

   Documented, plausible, open, contested. Attach the label to the claim and keep it attached when the claim is quoted elsewhere.

6. Step 6

   Date the conclusion

   State a conclusion as held on a date and on stated grounds. A conclusion that has been overtaken by measurement was still honestly held; erasing it destroys the only evidence that the method works.

What stays undelegated

Drafting can be shared. Searching can be shared. Structuring can be shared. These four cannot, and the reason is not sentimental: they are the points at which consequence attaches to a person.

• The decision about what the work is for.
• The decision to publish, send, or act.
• The judgement of whether a source is adequate.
• The acceptance of consequence when the judgement was wrong.

Stating a conclusion that may be overtaken

The interval will end. Some claim now labelled open will be settled, and some position taken in this volume will turn out to have been wrong. The procedure is built so that this costs the record nothing: the claim carries its standing, the conclusion carries its date, and the transcript shows the grounds.

What the procedure refuses is retroactive tidying — quietly revising a held position so that the author appears never to have held it. That practice destroys exactly the information a later reader needs.

Closing the volume

The sorting rule lost its empirical monopoly. The binary survived, but stopped being free. The framework that replaces it keeps two claims apart: content judged on merit, and responsibility kept human. Everything in this last chapter is that second claim, written out as steps.

The machine is a participant in the process of thinking. The governing faculty stays with the human — not because the machine cannot think, but because only one party to the exchange can be held to it.