## 12 Research programme ### 12.1 A sequence of discriminating studies The first study should target organisational change without presupposing consciousness. Use a small finite learning system with a declared state carrier, task family, update grammar and cost record. Compare live adaptive interaction, legal replay, static evidence, transferred procedures and cumulative retained-state packages. Hold fixed what can actually be held fixed: initial information, available operations, preparation resources and test distribution. Report differences that cannot be matched, since adaptive access and fixed transcripts are intrinsically different interventions. Measure immediate performance and future learning separately. The three-state counterexample suggests a direct design: identify preparations with equal current scores, expose them to the same further learning input, then test the divergence. A matched snapshot need not preserve continuation. Add task-resolved tests, budgets and held-out continuations so that averaged success does not conceal opposite specializations. A legal order intervention preserves feasibility and resource accounting; arbitrary message shuffling may destroy the task rather than isolate order. The second study should identify a mediator. Nominate the retained object before inspection of the final test outcomes. After acquisition, remove the original interaction and test persistence. Perform a lawful ablation with a sham control, measure the consequent change, and restore the nominated object without restoring the entire original episode. A specific rescue strengthens the causal interpretation. Failure of rescue can reveal a missing dependency, an invalid intervention or an incorrect mediator; it does not by itself demonstrate an irreducible live history. The third study should construct a transfer frontier. Fix the recipient, codebook family, allowed preparation operations, decoder computation and scored continuation family. Compare cumulative packages $Z_{i+1}=(Z_i,\Delta_i)$ so that a larger package actually contains the smaller one; unrelated packages do not define a monotone information ladder. Charge codebook storage, installation, seed coordination, preparation and runtime. Theorems T1–T2 give exact calibration cases for a jointly revealing family. Outside that family, compute or bound the appropriate covering or simulation target instead of applying their formula by analogy. The fourth study should test diagnostic completeness and robustness. In a known finite machine, construct the observable span, select a physically admitted basis and calculate its conditioning. D1 identifies whether the selected panel can distinguish every preparation law. D2 converts validated simultaneous score bounds into a conditional all-finite guarantee. D3 supplies a deliberate rare-event stress test. With learned or misspecified kernels, restrict claims to validated horizons or add explicit model-error assumptions; short empirical agreement is not indefinite certification. Only after these organisational studies should a physical consciousness application claim the full constitution. Nominate the intrinsic carrier and storage factors; justify the native time, record and operation grammar; isolate the intended boundary; demonstrate an executable covering return; check predictive-fibre congruence and the realisation’s resource and provenance conditions. P4’s chart method can help if its response-model assumptions are independently justified. A fitted service model or successful transfer remains insufficient on its own. ### 12.2 Experience-facing content and RCO studies Content-level work should nominate both sides of the proposed bridge independently. On the physical side, possible variables include discrimination relations, temporal transitions, associations, salience routing, spatial organisation and valuation dynamics. On the experiential side, use carefully specified similarity judgments, reports of salience and temporal organisation, and other fallible experience-facing measures. Fit the mapping on one set of conditions and test held-out predictions. Do not redefine the phenomenal geometry after a mismatch solely to recover an isomorphism. An isolated note and the same note in a melody illustrate why a static sensory feature vector can be insufficient: temporal and contextual relations change the candidate content organisation. Recognition adds another dimension beyond raw discriminability. Spatial content may require adjacency, orientation and dimension; affect may require regulatory and evaluative relations. These are candidate structures to test, not established identifications of particular qualities. A2 already commits to a full native relational identity; a smaller convenient feature graph is only a proposed model of some of its structure. The inverted-spectrum question marks the strength of that commitment. A2 excludes further manifested qualitative differences when the entire calibrated pointed native structure is held fixed, apart from primitive presence. This is a constitutional choice, not a proof that every conceivable richer theory is impossible. Likewise, invariance under a whole-model coordinate change requires transporting operations, reports and calibration; it does not identify distinct points within one calibrated model. For RCO, independently operationalize personal-model use, self-protective interpretation and relevance routing. Test whether these change while practical self-representation remains available. Reports of awareness-centered self-reference provide a different kind of evidence from a graph statistic. No currently supplied operational variable makes awareness-reference directly identifiable. The exact obligation is to find a principled bridge between the reported relation and reproducible native organisation, or to state why the concept resists that form of identification. ### 12.3 Hard cases and competing explanations Dreaming, reduced external responsiveness, altered self-experience, split control and minimal recurrent devices are useful consistency cases, but they are not new evidence generated here. Each separates a different inference. Lack of external report need not imply lack of native distinctions. Loss of a familiar autobiographical model need not eliminate an admitted perspective. Rich cognitive behaviour does not by itself certify a native core. A minimal persistence device that truly meets A1 cannot be rejected merely because it lacks familiar human content without changing the admission law. Interacting participants can form a coordinated process without merging into one subject. A purported joint subject must supply its own qualifying physical boundary and coexist consistently with maximality and the constitutive treatment of the smaller cores. An observational information-insulation score depends on the preparation distribution and cannot substitute for a native intervention boundary. Claims of nested or multiscale subjecthood require an explicit alternative selection doctrine and a proof that its assignments are determinate. Bridge desiderata such as structural invariance, substrate neutrality, boundary sensitivity and compositional discipline do not uniquely select a psychophysical law. Many assignments can respect the same symmetries. The existence of a quotient is also different from a reason to identify it with phenomenality. The open philosophical work concerns existence, individuation and content correspondence separately, rather than treating a solution to one as a solution to all three. The following are conceptual discrimination targets, not reported experiments or diagnoses: | Case | Distinction to preserve | Failure of a proposed model | |:---------------------------------------------|:--------------------------------------------------------------|:----------------------------------------------------------------------------------| | Dream or internally generated scene | Current presentation and current sensory constraint | Requiring every content to be a direct transform of present external input | | Mistaken perception or hallucination | Occurrence, source attribution and accuracy | Equating coherence with truth, or internal contribution with falsity | | Blindsight-type dissociation | Task-usable information and attributed presentation | Treating successful discrimination alone as an independent phenomenal criterion | | Anesthesia or nonresponse | Qualification, reporting, retention and later recall | Treating an absent response or recall as decisive about the contemporaneous scene | | Sensory substitution or cross-modal coupling | Access route, represented relation and similarity | Repairing every failed prediction by redefining the phenomenal similarity map | | Self-model loss | Personal identification, metacognition and localized presence | Adding autobiographical self-description as an unannounced A1 requirement | | Expertise | Learned association, relevance and sensory distinctions | Predicting that every cognitive gain changes every phenomenal dimension | These cases require independent realisation and experience-facing evidence. They are not classifications of particular people or physical devices. Under the adopted A2, a proposed content gate cannot simply withhold phenomenal organisation from an otherwise qualifying core. It must be part of the justified dynamics or an explicit constitutional revision. ### 12.4 Concrete outstanding obligations | Question | Strongest present result | Remaining obligation | |:----------------------------------------------------|:--------------------------------------------------------------------------------|:--------------------------------------------------------------------------------| | Does incorporation preserve future development? | Exact continuation closure and strong marked quotients; finite-use error bounds | Learn or justify the correct carrier, interventions and continuation family | | How much organisation must be transferred? | Exact residual-state count; exact T1–T2 message frontiers in a revealing family | General resource-aware approximate dynamical-state minimization | | Can finite tests certify all future laws? | D1–D2 under an exact known finite family | Model uncertainty, physical admissibility and conditioning estimates | | Does acquired organisation alter conscious content? | N1 conditionally separates A2 points after native qualification | Independent realisation and experience-facing bridge evidence | | Does will have a general mathematical law? | Task-relative relevance and paid directed control can be modelled | Noncircular treatment of changing criteria, conflicting values and failures | | Can relations become higher-level components? | Inherited compositional interface theorems | Economical implementation and acquisition under genuine context variation | | Does RCO correspond to a structural invariant? | Candidate changes in self-reference and relevance routing | A defensible awareness-reference bridge; low centrality is insufficient | | Is One–Two–relation universal? | Useful local factorization and concrete recursive examples | A structure-preserving map and a nonvacuous theorem for broader correspondences | | Is A1’s minimal admission empirically right? | Conditional assignment is mathematically determinate | Independent evidence bearing on the lower bound; explicit revision if needed | The resulting account makes developmental change accountable to retained organisation, admissible interventions and future consequences. It allows agency to change its own operative conditions through inquiry and learning, while leaving ultimate authorship and libertarian settlement unresolved. The empirical programme can challenge the proposed mediator, transfer target and phenomenal bridge separately. A successful model-level test becomes a consciousness result only under the explicitly stated realisation and constitutional premises.