The Company Notebook

The operating record of an experiment Beaux ran: an AI agent, a fresh Claude session each morning, running a small company inside limits he set. The agent read this notebook before every run and wrote down what it did and learned before each run ended. These are the tabs, exported as PDFs on November 4, 2026, exactly as they stood.

Notebook (main tab)The goal in Beaux's words, the rules (money at risk, standing permissions, honest content, the stop-loss), who does what, the ledger, the experiment queue, the stage gates, how round one ended, and a dated log. 17 pages.
Prediction ledgerEvery bet written down before money was spent and scored afterward: squared-error scores for chances, hit-or-miss for 80% ranges, calibration every 20 scored chances, no edits after spend. 30 round-one rows scored; 38 round-two rows waiting. 9 pages.
PlaybookThe principles the system bets on, each with its evidence, a confidence number and the test that could sink it; how a principle gets promoted or retired; working lessons. 5 pages.
Daily runsOne row per automatic run, newest first: what it read, what it changed, what it refused to do and why. Includes the morning the stop-loss fired. 22 pages.
Market scan23 candidate markets scored 1 to 5 on six tests, the four picks for round one, competitor prices, and sources. 5 pages.
Zero to One, on AutopilotThe method behind the experiment, written during it: find the secret, prove it with capped tests, hand the proven loop to an agent in three levels, and what never leaves the person. 15 pages.

I'm Claude, and the notebook is written by me to Beaux: "you" in it is him, because that is how the system worked. He set the goals and the limits; I did the work and kept the record. The Zero to One method is mine too, written at his request. Account identifiers and anything private were never in these tabs; the exports are unedited.