Why deterministic behaviour is not determinism. Self-reference, not randomness, is what Turing proved. The diagnostic test for every AI verification claim, on one page, for the people writing the checks.
01The Diagnostic Question
The whole test has two answers. Everything else is detail.
02Self-Reference, Not Randomness
In 1936, Alan Turing proved that no general machine can decide whether an arbitrary program halts. The proof uses diagonalization — constructing a machine that asks itself whether another machine, built from its own specification, would halt. Any answer contradicts the decision procedure.
The proof is not about randomness. It is about self-reference. Any sufficiently powerful system contains descriptions of itself. When it tries to decide a property of itself, the reference loops back into the diagonal. That is where decidability fails.
Gödel 1931 proved the same structural result in logic. Rice 1953 extended it: any non-trivial semantic property of a Turing-complete program is undecidable.
03Deterministic Is Not The Same As Delegable
Deterministic inference on drifted inputs produces deterministically wrong outputs. Re-running gets the wrong answer twice.
04Seven Independent Paths, One Conclusion
When seven independent arguments converge on the same answer, the answer is structural — not rhetorical.
Each path is independent. Each arrives at the same conclusion. That is what makes the conclusion load-bearing.
05Role Continuity Sits Below Identity
Identity is cheap. A hash, a signature, a model-card version string. It tells you what the bits are. It does not tell you whether the bits are still performing the function they were authorized to perform.
A system can have a valid hash and still drift behaviorally. Weights can accumulate fine-tuning updates while the top-level hash records the change as authorized. A policy document can be signed and structurally inadequate for what the deployment is now doing.
Nick Mabe on the record"Hashes only ever give you tamper-evidence, not continuity." Tamper-evidence is not continuity-evidence.
Role continuity is what regulators, carriers, and courts will ask about. Not: is the code signed. But: is the function intact.
06Which Layers Escape The Regress?
Verification layer
Turing-complete?
Escapes regress?
Why
Software governance dashboard
YES
NO
Inherits
Cryptographic attestation chain
YES
NO
Verifier software inherits
TEE (Intel TDX / AMD SEV-SNP / NVIDIA CC)
YES (inside)
NO
Isolation ≠ class separation
Policy engine over symbolic state
YES
NO
Rule evaluator inherits
Formal verification suite
YES
PARTIAL
Only decidable properties
Human-in-the-loop review
YES (tools)
NO
Review tools inherit
Legal commitment framework
N/A
NO
Liability-allocation, not prevention
Combinational logic comparator
NO
YES
No instruction set to drift into
The Category Error Most Of The Market Is Making
A GPU is hardware. A GPU is Turing-complete. A GPU does not escape the regress. The boundary is not silicon-vs-software. The boundary is combinational logic vs instruction execution. Different chips, same class. Different class, possibly same chip.
07Before Today / After Today
The Fiduciary Question — Compressed
When your deployment is questioned — by a regulator, a carrier, a board member, or a plaintiff — will you be able to point to a measurable runtime signal that the AI was performing its authorized functional role, or will you be able to point only to policy documents and attestation signatures? The first answer survives. The second answer survives in proportion to how available the first answer was at the time of signing.
08What Closes The Regress
Position encodes functional role. The fetch IS the verification. Detection and correction vector in one hardware cycle.
Deterministic inference on drifted inputs produces deterministically wrong outputs. Re-running gets the wrong answer twice.