Relational development Section 6
Published constitution and realisation
6 Published constitution and realisation
6.1 A0–A3 and the scope of the developmental proposal
A0 identifies awareness with the knowing aspect of reality prior to the operational source/readout distinction. It does not introduce a localized subject, an extra force, or a numerical global experiencer. The finite physical and predictive constructions can be studied without deciding whether they manifest awareness. A0 supplies their ontological interpretation.
A1 assigns one localized perspective to a qualifying maximal internal strongly connected component. Qualification requires an executable covering return and at least two distinct native predictive classes in the actual operating type and resource context. The covering schedule must be physically admitted and must preserve a distinguishable return through the nominated core. A cycle made from mutually incompatible control contexts is insufficient. Overlapping recurrent subassemblies inside one admitted maximal core do not acquire additional subjects merely by being graph cycles.
This law is deliberately permissive. A correctly realised one-bit persistence process can qualify. Language, intelligence, explicit self-modelling and rich relevance arbitration are not additional A1 requirements. The richer scaffold developed here describes an important subclass and a developmental interpretation. Excluding every minimal memory, admitting nested overlapping subjects, or requiring a world-for-the-vessel as a new necessary condition would revise the constitution. Such revisions may be worth investigating, but must be named and defended as revisions.
A2 identifies phenomenal relational organisation with a structural copy of the complete endogenous predictive object over all physically admitted finite continuations. It preserves native operations, outcome labels, test laws, horizon restrictions and well-defined class transition instruments. The actual phenomenal point is the image of the predictive class of the realised intrinsic state. A metric summarizing maximum distinguishability is not the complete object. Different organisations can have the same coarse distance pattern.
A3 supplies episode continuation through physical process provenance and unique continuation of qualifying cores. Information copying and process continuation are distinct relations. Material renewal can preserve a documented operative lineage, while a newly allocated functional copy can begin a new admission. Genuine gaps and the specified split or merge cases terminate the relevant old episodes under the adopted law. Neither preservation of primitive awareness nor continuity of a personal narrative proves numerical identity of an experiential episode.
These are constitutive commitments. Mathematical determinacy conditional on them does not derive them from an independently established theory of matter. This distinction is needed for constructive criticism: a disagreement about A1’s lower bound is different from a failed physical recurrence certificate, and both are different from an inconsistency in quotient construction.
6.2 State, law, predictive history and capability
Several objects must remain separate throughout a developmental argument:
| Object | What it retains | What it does not automatically establish |
|---|---|---|
| Complete realised state | Variables needed for the nominated continuation law | Physical accessibility to a learner or encoder |
| Actual native predictive class | Equality of all admitted finite native transcript laws from actual states | Strong successor-class congruence in a general hidden stochastic machine |
| Predictive history state | Conditional continuation law after an observed history | Identity of the actual hidden state |
| Strong marked quotient | Joint records, costs and actual successor classes | Minimality among all possible stochastic generators |
| Capability profile | Selected task scores and resource-indexed success | All future learning or native prediction |
| Physical realisation | Boundary, factorization, interventions, native timing, resources and provenance | Empirical truth of the phenomenal identification |
Under a fixed complete carrier, installed programs, memories and valuation parameters belong to the state. Learning then ordinarily moves the actual point inside one unpointed predictive object. A claim that the whole object changed requires a change in the nominated laws or contract, or a justified comparison between different realisations. Renaming a state-dependent program as a new law does not create this stronger conclusion.
The distinction also limits content claims. A change in a grounded report law can imply different A2 classes when grounding is established. An arbitrary laboratory observation need not descend through the native quotient. External apparatus that leaves the intrinsic state and native contract unchanged cannot change the phenomenal object merely by measuring it. Apparatus that supplies feedback, drains resources or changes the boundary can of course change the premises.
6.3 Why realisation cannot be inferred from service behaviour
P4 distinguishes trace conjugacy from preservation of the physically nominated intervention algebra. A recoding can preserve complete decoded service traces while mapping a one-coordinate overwrite to a multicoordinate operation. Two descriptions of one transported realisation must carry the intervention translation with them. Two different implementations that each call their own coordinates elementary may have different constitutive inputs even when their services agree.
The positive P4 identification result is conditional. With a fully distinguished finite binary-state carrier, positive product response laws and sufficient rank, the binary chart can be recovered up to signed coordinate permutation. Its robust version requires the stated spectral separation and error conditions. These results can support a developmental register experiment when their premises hold. They do not establish that a physical preparation actually obeys the product-noise model, identify every missing realisation field, or validate consciousness from chart recovery.
P3 supplies a complementary learning lesson. Representability, candidate production, model selection and intervention validity are different achievements. Its supplied source reports that the specified semantic family covers 48 of 48 benchmark items, while smaller fitting produces 39 of 48 good candidates and selection retains 37 of 48. Those are source-reported results, not rerun here. An unsuccessful selected model need not reveal an inadequate representation language, and a larger fitting budget need not improve every selection. The new programme should retain these distinctions when testing learned scaffolds.