The protocol.

Six operating principles, derived from what worked and what did not.

1. The curriculum precedes the research. We did not attempt research until the curriculum was complete. Thirty-eight documents, in order, with each document closing before the next opened. The temptation to skip ahead is real. We did not skip. The curriculum is load-bearing.

2. The human sets the direction. Every decision about what to study next, what problem to pursue, what result to include, was Adam’s. Claude executes with exceptional speed and precision. Claude does not pick the problem. The human picks the problem.

3. Calibration is explicit. Before each research phase, we stated a probability estimate for the expected outcome. After each phase, we recorded the actual outcome. This is the calibration log. It exists so that the project cannot be retold as a story of inevitable success.

4. Mistakes are documented, not hidden. Several predictions were wrong. Several approaches failed. These are in the record. The phase memos are public exactly because they show the wrong turns.

5. The credit string is non-negotiable. “Claude (Anthropic), computational collaborator.” Not assistant. Not tool. Not co-author. Collaborator. The distinction matters and we hold to it.

6. The chat is private, the artifacts are public. The conversations themselves are private. The outputs of those conversations, which is to say the curriculum documents, the phase memos, the paper, and this site, are public. The transparency story rests on the artifacts and their dates, which are self-evidencing.