Is the ARC Theory science, or only a hypothesis?
The ARC Theory states five laws under test, with published refutation conditions. The programme's own Unit 83 run supports P21 on its registered access axis: family-dependent, with its mechanism axis not evaluable and a disclosed recovery amendment. This is a scoped result within a research programme; all five laws remain under test.
Judge it by its definitions, its rivals and its decision rules. Each symbol of Laws I to III has an operational definition paired with the measurement that would produce it, and each proposition carries the observation that would refute it, fixed before its data. The registration names rival explanations: for P21 it states the honest expectation that a rival may explain part of any widening found, possibly most of it. Unit 06 is administratively incomplete, with no verdict, and earlier capability and alignment studies carry their own corrections and limits, the retraction of the 2.24 estimate among them.
The theory has working tools and a first registered result beside it: Eden Control and Eden Legal AI, both in private beta; the ARC Evaluation Protocol, in development; and Unit 83, the first registered study to reach a verdict. A public registration and a corrections record make the claims accountable; they do not replace sound measurement, independent criticism or replication.