The claim
The owner, at U15, after the dynamics of transitions (QUESTION-281): “inquiry is a form of specification.”
A specification constrains what can count as an admissible realization (CLAIM-477). A question does the same for answers. A15: “A question is therefore not merely a request for information about an independently specified object. It can be a partial specification of the space in which an answer—and sometimes the object of inquiry itself—can become determinate.” Formally, a question q: Ω → A_Q partitions states into those it cannot tell apart. An experiment adds interventions and observations, so A15 wrote an inquiry as (A, 𝓘, 𝓞, 𝓚): answer space, admissible interventions, observations and criteria of evaluation. A16 added the language Σ and the resource bounds 𝓑. Neither called this a standard definition.
Three parts, with what supports each:
- A question is a partial specification. “Is this system stable?” does not yet say which system, which perturbations, which stability or which horizon (A16 §1). Belnap and Steel’s logic of questions and van Fraassen’s contrast classes (LIT-1637, Unavailable) make the point formally; A16 cited both, and neither is read here.
- A finding can stabilize into a specification. A result reliable enough becomes a theory, method, instrument or rule that governs the next inquiry (A15 §3). Collingwood’s logic of question and answer (LIT-1651) is the epistemic form: an answer established once becomes a presupposition of the next question. Rheinberger’s epistemic things becoming technical objects (LIT-1665) is the experimental form.
- An inquiry can revise its answer space. This is where inquiry is more than a specification. A16 §4: a question about which equilibrium a system occupies has no adequate answer once a Hopf bifurcation leaves a stable limit cycle, and must become a question about which attracting set organizes the dynamics. Dewey (LIT-1645) puts the formulation of the problem inside inquiry.
What the exchange added
TERM-065 already holds half of it. The effective-description regime ℛ = (X, 𝒬, ℐ, ℬ, ε, τ) keeps a candidate description apart from the conditions it is judged under, so descriptions can be compared without moving the goalposts. A16 §2 drew the second operation from it: “Evaluate T under ℛ” and “Evaluate or revise ℛ”. “Inquiry becomes reflexive when its own conditions of evaluation become objects of inquiry.” TERM-065 treats ℛ as supplied; this claim says ℛ is itself made by inquiry and can be remade by it.
The formal case is counterexample-guided abstraction refinement (LIT-1652). A model checker runs on an abstraction that merges states. When a counterexample proves spurious, the abstraction is refined to separate the states whose merging produced it. A16 §6 read luria’s DP-016 (CLAIM-419 here) as “the same epistemic pattern”: failure under a specification, diagnosis of a missing distinction, refinement. It also gave the limit: CEGAR refines the abstraction for a fixed property, and “does not automatically recognize that the verification property itself is misguided.” That is the line between refining within a framework and revising it, and the claim’s last clause is about the second.
Precision is not correctness. Expected information gain (Lindley, LIT-1675) is relative to the hypothesis space already specified. A16 §7: “The inquiry can become increasingly precise without becoming correct.” The record has the empirical form in Devezer et al. (LIT-988, read in NOTE-792): sound model comparisons need not lead a community to the true model, because which alternatives are proposed constrains what it can discover. A15 §6: “So specification is not epistemically innocent.” TERM-056’s question drift is a case of it. A15 drew a criterion from it, with Chang’s pragmatist realism (LIT-1641) beside it: a healthy apparatus “must support not only the evaluation of admissible answers, but the detection and revision of inadequate conditions of admissibility.”
Two kinds of failure. A16 §10 separated a failure of realization (the inquiry is adequate and this attempt failed) from a failure of specification (the right kind of answer cannot be expressed). “the same observed failure can demand entirely different responses”, and deciding which it was is itself inquiry. A16 §14 counted three things an inquiry can find: an answer it did not know, a distinction it did not know how to make, and a limit of the practice by which it decides what is knowable.
What it does not claim
- That inquiry makes what it investigates. A16 §9: “This doesn’t imply every law of nature is produced by scientific inquiry. It means the epistemic specification through which we investigate a law can be historically constituted.”
- That inquiries undergo bifurcations. A15’s table pairing anomalies with instabilities and paradigm entrenchment with persistent organization is “a structural comparison, not an assertion that inquiries literally undergo physical pitchfork bifurcations”. Whether the dynamics of inquiry can be modelled with the same mechanisms is open.
- That the regime should float while descriptions are compared. Keeping ℛ fixed for a comparison is a local methodological commitment (A17 §8), and this claim does not loosen TERM-065.
Why it bears on the record
CLAIM-476 says luria is an apparatus for a record’s objects whose grammar the record’s own inquiry revises. This claim is the general form: any inquiry specifies, and its stable products are specifications. A16 §11 read luria’s split between editing a record and revising its schema as “working within an inquiry regime and investigating the regime itself”. CASE-178 measured the second in this record: it happens rarely, on demand and by decision. CLAIM-475, the owner’s next turn, adds that specifying is itself bounded by resources.
What would test it
A16 §12 proposed one study. A synthetic system can be stationary or oscillatory, and investigators start with a schema that knows only kinds of equilibrium. Three regimes are compared on the same external benchmarks:
- fixed (update answers only);
- refinable (add distinctions within the ontology);
- revisable (change questions, variables and evaluation). If the revisable regime finds the oscillation no sooner and no more reliably than the refinable one, the last clause loses its point. The common benchmarks are what keep the comparison from rewarding moved goalposts.
Note of 2026-10-27 (evening): the readings of the U15–U23 works
- CEGAR (NOTE-1231; THEORY-1064, THEORY-1094) supports the claim’s first part. The property’s atoms build the initial abstraction, and no abstraction may merge states that disagree on them (Thm 1, p. 5), so the question fixes which differences may never be merged. Its loop has the exact shape of defeater (a). Every refinement is a split between the first abstraction and the concrete state space, and nothing touches the property. CEGAR is therefore a worked case of refining within a fixed answer space, as the body said. It does not count against the claim’s last clause, which is about inquiry in general. The coarsest split that removes a counterexample is NP-hard to find (Thm 3).
- Simon (NOTE-1232) gives a partial case of the last clause. Chess “involves continually redefining what the problem is” (p. 186). In design, requirements are evoked mid-process and need not be specified in advance (pp. 190–191). But most of his revision is selection, “generated from, or dredged from” memory (p. 188), which leans toward defeater (a).
Note of 2026-10-27: the term
The owner asked the same day for a TERM for inquiry, “an important building block of our formal argument”. TERM-095 defines it as the tuple (Σ, A, 𝓘, 𝓞, 𝓚, 𝓑) used above, and leaves this claim’s thesis out of the definition. So the claim is stated in the term and not true by definition.