Friction check: Irreducibility to rules bounds text alone, not text plus an interpreter

Note: kb/notes/irreducibility-to-rules-bounds-text-alone-not-text-plus-interpreter.md Central claim (one sentence): An irreducibility-to-rules argument excludes retained text from deciding novel cases unaided but does not exclude a composite in which the text preserves premises and a competent interpreter exercises the unreduced judgment, unless a further premise rules out such an interpreter.

Filter

Verdict: SURVIVES

Rule-irreducibility, premise retention in text, and case-by-case judgment by a separate interpreter can coexist if the interpreter genuinely supplies the capacity that the text cannot reduce to rules and “text alone” means text paired only with execution machinery that adds no such judgment. This witness leaves the existence and competence of an LLM-based interpreter as a separate conjecture rather than part of the logical distinction.

Signal — thinnest joints

  1. “So the thesis that program theory resists externalization is really a thesis about one of its parts, and the parts have different substrates.” — UNSUPPORTED — The preceding observation that reasons and fixed mapping points can be written down establishes that records can serve as inputs to a reader; it does not establish that those records perform Naur's justification and mapping capacities. Those capacities may require interpretation just as novel similarity judgment does, so the claimed capability-by-capability substrate split has not been derived.
  2. “Its counterfactual character points the same way” — UNSUPPORTED — The existence of unmaterialized alternative programs in a programmer's understanding does not distinguish an irreducible application-judgment capacity from a propositional or rule-governed representation: finite models and procedures can also represent, generate, or compare counterfactual alternatives. The counterfactual observation therefore supplies no stated bridge to the claimed representational conclusion.
  3. “In 1985 the identification needed no defence, because the set of candidate world-knowing interpreters that could sit inside a software system was empty. The negative half of the thesis ... is therefore a report on the available substrates” — UNSUPPORTED — Historical absence can explain why Naur did not discuss a software interpreter, but it cannot establish that his human-only conclusion was contingent on that absence rather than on an unstated claim about understanding, participation, or personhood. The note's source caveat preserves this possibility, so “therefore a report” outruns the evidence it names.
  4. “Nothing in the first clause selects a human bearer; it selects any bearer that understands the world.” — THIN — The quoted clause says “someone,” and the note does not establish that Naur used understands as a substrate-neutral functional predicate. Separating the requirement from the named programmer exposes a possible extra premise, but it does not by itself license replacing a person with any bearer that produces the relevant judgments.
  5. “A capacity of that shape cannot be discharged by a rule set, and a text that must decide unaided is a rule set.” — THIN — Naur's quoted irreducibility claim supports the first half, but the second needs a concrete account of what it means for inert text to decide and which execution machinery is included in “unaided.” Without that boundary, the reason does not establish that every text-based decider is a limited criterion set while the proposed text-plus-interpreter composite is not.

For the human

Look first at whether written justifications and fixed mapping points actually relocate Naur's corresponding capacities into text or merely provide more premises to an interpreter that still must hold all three capacities.