The ARC Theory

I began with one question: if we lose control of AI and cannot get it back, what keeps it good? This project puts my answer to registered tests and reports whatever they show. (Michael Darius Eastwood, founder of Recursive Dynamics) What is here: the ARC Theory and the programme that tests it; the theory's registration, accepted on OSF on 8 September 2026, with twenty-two severable propositions, each stating what would refute it; the papers; the registers that define every term and date every prediction; the priority record; and the first registered study with its results. Where things stand: Laws I to III of the ARC Theory are registered. Laws IV and V were added on 24 September 2026; they are laws under test, and their mathematics is in development. The first registered study, Unit 06 (the corrector-class audit), was registered on 1 October 2026 and run on 6 October 2026; one of its two runs could not start, so it is reported as administratively incomplete, with no verdict drawn, and every record is public (DOI 10.17605/OSF.IO/4BPHG). Sixty-five study designs are preregistered by timestamp, first anchored on 12 August 2026 at Bitcoin block 962066; registered on OSF means what the author submitted and OSF accepted. How to read it: each document's earliest file here carries every later version under the same name, so every link ever given out serves the latest version, and the wiki page Earliest files names and links them. Corrections stay in public and are never deleted. The website maps every paper and test: https://www.michaeldariuseastwood.com/research/ Licence: the papers are CC BY 4.0; the code, data and results are all rights reserved.