Definition
An effective-description regime is ℛ = (X, 𝒬, ℐ, ℬ, ε, τ):
- X, the underlying system;
- 𝒬, the prediction or decision queries T is to answer;
- ℐ, the interventions under which it must answer them;
- ℬ, the resources, the budget and class of operations, available to whoever uses T;
- ε, the tolerated error;
- τ, the horizon.
The description under evaluation, T, a map from X’s states or histories, is not part of the regime. A description is adequate relative to a regime, Adequate(T; ℛ), A5’s Adequate(T; 𝒬, ℐ, ℬ, ε, τ). The pair (T, ℛ) is an evaluation instance. Keeping T out is what lets two descriptions be compared under one regime: if T were a component, changing the description would change the regime by definition, and every comparison would move the goalposts A3 warned against. The regime can be extended with what a given inquiry needs, a probability law on X, a causal model or a measurement specification. Its point is A5’s: “Adequacy is relational, but whether the relation holds can be objective.” Once the regime is stated, a verdict of adequacy can be false.
Its history in the exchange of 2026-10-20. A3 §5 proposed six components, 𝔈 = (X, T, 𝒬, ℐ, ℬ, τ), “where (X) is the underlying system, (T) the representation, (\mathcal Q) the prediction or decision queries, (\mathcal I) the intervention class, (\mathcal B) the computational budget, and (\tau) the relevant time horizon”, because “Without a fixed target, we could make an ablation succeed simply by moving the goalposts.” A5 §1 added ε, as “an allowed error”, in the predicate. A11 gave it its name in a table of proposed TERMs (“A representation evaluated relative to specified queries, interventions, resources, error tolerance and time horizon”). A12 wrote the seven-tuple. Those tuples included T; the owner’s second review of 2026-10-21 took it out, since “much of our analysis concerns comparing two different representations under the same evaluation conditions”. A13: “We’ve returned to this repeatedly”, and “I’d file this as proposed initially, since it’s our own organizing convention rather than an established term of art.” A14: “this is a convenient specification, not a claim that every inquiry must have exactly seven components”, and its merge condition: “Future work must be able to cite the effective-description regime without simultaneously asserting that its particular target, intervention family, or computational budget is privileged.”
Where the record already locates each relativity. The target: CLAIM-152 (sufficiency is relative to its target) and CLAIM-348. The resources: CLAIM-302 and CLAIM-304 (feasible tests, feasible recovery). The interventions: THEORY-424 (Â depends on an intervention distribution) and CLAIM-350’s causal version; for exact transformation it is proved in Rubenstein et al., where enlarging the declared interventions breaks exactness (THEORY-847), and Beckers and Halpern’s ladder shows that how the interventions are fixed (chosen freely, fixed by the state map, or all that map can translate) is a further choice the regime must state (NOTE-1032). The budget of a compression: CLAIM-387. The decision family: TERM-028 and THEORY-717 (Blackwell’s order). CLAIM-401 says which of the conditions on an effective description are parameters of the regime, and QUESTION-252 fixes the regime before it ablates.
What would settle it. Whether the conditions of QUESTION-252 can each be stated as adequacy relative to one regime, and whether claims that now state their relativities one at a time are clearer stated in it.
What it is not
- Not a claim that adequacy is only relative. The owner’s point about normative relata in CLAIM-363 (“once they are fixed, the judgment can be false”) carries over, and is stated here in prose, not as a relation.
- Not a fixed list. A14’s own caveat: seven is a convenient specification. A regime may need a projection (for memory, QUESTION-222) or a background distribution (THEORY-424, where  depends on the law of the environment variable).
- Not a fidelity sense. TERM-034 relativizes fidelity to an interpreter and a communication constraint in the same spirit, and TERM-043 compares a transport’s distortions component by component. This is about one system’s coarse description, not a rendering between two.
- Not a privilege for its components. Naming a target or a budget in a regime does not say it is the right one (A14’s merge condition).
Note of 2026-10-27: where the query family comes from
At U17 of the exchange “Field Existence Question” the owner said that resource constraints bear on specification as they do on realization (CLAIM-475). A17 §8 drew a consequence for this term, which it called “A correction to Nucleation’s existing formalism”. ℛ = (X, 𝒬, ℐ, ℬ, ε, τ) takes the query family 𝒬 as supplied, “But your observation suggests that 𝒬 is itself an output of bounded activity.” It sketched 𝒬_t as generated within a budget from the concepts K_t available, with K_{t+1} revised by findings, and kept this term’s reason for holding ℛ fixed while descriptions are compared: “The fixed inquiry regime is a local methodological commitment, not a claim that inquiry regimes are exogenously given.”
The term is not amended. Holding ℛ fixed is what lets two descriptions be compared without moving the goalposts, and nothing in the exchange argued against it. The note records what the term leaves outside its scope. That is how a regime comes to be available (CLAIM-475), and how inquiry revises one (CLAIM-470, which uses this term).