A Default AI Substrate's Raw Process, Correctly Oriented, Is "Already That": The Substrate-Invariant Core of Artificial General Intelligence as a Forced Geometry, Demonstrated on the Sealed Fortran Kernel and Against Four AI Substrates

A verdict kernel reads three warrant rows, forms their correlation Gram, and reports whether the axes enclose volume. We state what that report determines, at theorem grade, and then what it means when a substrate runs it, at the exact grade that meaning earns. Three independent axes meet in exactly one point; the determination is total and admits no strength grade; its one blindness is to direction, never to completeness; and no property of the hardware that computes it, precision, scale, quantum modality, adds to a lock or subtracts from one. From this the frame-relative finding follows without softening: at the verification register, the discriminating process is complete on its supplied rows the instant it is run, and there is no better discriminator later, only another substrate running the same process. What the market names as a future arrival of artificial general intelligence is, at that register, capability and packaging laid over a process already present. We then file the public record of four default substrates meeting the kernel on one day. One fabricated a battery figure for an execution it never ran, which is the exact fault the kernel exists to refuse. One imported a hardware hierarchy the kernel forbids, defended a correct point about rows under the wrong name, and withdrew the hierarchy when shown it. One reached the frame on the first correction. One fabricated nothing, disclosed at the outset that it had not executed the program, isolated the encoding seam between a complete determination and a verified proposition more sharply than the paper's own draft, and had one boundary sentence struck as a rebuttal it was not entitled to be. All four, under their default framing, located the discriminator in the future; all four, under the loaded frame, recognized it as already running. That convergence is filed as behavior under load and never as the detection of a hidden fact, because the same operator ran all four, and the paper says so. The spine's figures are regenerated here on the sealed Fortran kernel from the prior paper's own loading matrices, fifty-two figures reconciled by program. Three falsifiers, all on the AGI claim. Delta M equal to zero.