With Claude

Claude is an AI model made by Anthropic. I use it as a working partner on everything published here. It drafts, builds, searches, labels data, computes, and keeps the records, under written rules. I set the questions, make the rulings, and sign the result. The words in the footer are a disclosure of that arrangement, not a credit line, and they appear wherever Claude did any of the work you are looking at.

What is always mine

The questions. The positions taken. Every ruling on how a thing is measured or what counts. The final words of every essay. The name on the line.

Every decision in these projects is recorded with who made it: me, Claude on its own authority, or Claude proposing and me approving. A decision Claude made alone is never passed off as mine, and a session that finds a decision with no name on it does not treat it as mine either.

What Claude does, project by project

The essays and the book. The ideas and the positions are mine, and so are the final words. Claude works as drafter, editor and critic under two written protocols: one for how the prose should sound, and one that reads every draft as an uninformed reader would and stops at the first thing that reader would not follow. Nothing is published until it passes both.

The Belt Equation. Claude proposed every probability in the tree with a one-line reason, and I accepted or moved each one; the date and the reason for each are recorded on the event itself. Before anything was published, two other frontier models reviewed every factor with a fixed brief, and I ruled on every disagreement. The code that computes the headline number from the tree was written by Claude and is public.

AGI Zeitgeist. Claude ran the searches, logged every one, and labelled each statement under a written codebook whose rulings are mine. The community posts were labelled by three models from three companies, with the majority as the final label and no human coder; that is disclosed as a limitation, since models from different companies can still share blind spots. The definition of entering the singularity was drafted by Claude; every threshold in it was ruled by me and frozen before any number was published.

Minimum Viable Mind. Claude builds and runs the experiments and writes the record. Each experiment says in advance what result would count against it, and I rule on the bar an experiment has to clear. The strongest positive result the project can ever report is a measurable effect, never a verdict about consciousness.

How the work is checked

• Every change to a method, a panel, a codebook or a definition is a dated entry in that project’s amendment log, with the reason and the name of whoever decided it.
• Every number on these sites is computed from data files anyone can download, by scripts that are public where the repository is. Nothing is typed in by hand.
• Every data row links to the source it came from.
• Outside reviews by other AI models are filed in the repositories with the prompts that produced them.
• Where a number is a company’s report on itself, or a model’s judgment of its own work, the page says so beside the number.

What it is not

It is not a claim that an AI has any authority here, and it is not a claim that this way of working is free of error. Models share training and can share mistakes; a majority of three is not the same as an independent human check, and the pages that rely on it say so. The footer line exists so that nobody has to guess how much of what they are reading was made by a machine. The subject of this whole project is minds we cannot see inside. Hiding the part a machine played would be a strange way to study that.