What Would Test the Shadow Consciousness Model?

SPC-2 can be assessed by testing proposed physical realizations and their fixed correspondence to independently elicited phenomenal relations. A model can fail to predict records, select an unjustified boundary or mismatch experience-related structure. These are substantive failure possibilities. The finite completion theorem does not settle them.

There is also a precise limitation. If the entire physical realization and every report law are held fixed, adding the source-aspect interpretation changes no physical transcript probabilities. A transcript-only experiment cannot then distinguish that interpretation from an otherwise identical physical account that denies it.

Useful testing begins by naming the claim. Is the target a component model, a sufficient descriptor, subject individuation, the A2 correspondence or the source-awareness premise? A positive result about one must not quietly become validation of all the others.

The conservativity theorem and Chapter 23.5 establish this division.

A candidate realization must identify actual components, time scales, ports, preparation and intervention possibilities, resource updates and boundary conditions. Native operations cannot be whatever the analyst wishes were available. The model must account for memory that can return through an external interaction and for controller state that affects later operation.

Before assigning experience, test whether the model predicts the relevant physical records under held-out conditions. A fit to one preparation family is weaker than closure on every state in the nominated domain. Candidate projections that fail closure may need enlarged carriers or a history treatment. Failure of the representation is not yet a negative verdict about consciousness.

Realization selection is itself part of the research. The same bare transition kernel can admit different primitive factorizations, and a symmetry obstruction prevents a unique equivariant choice in a finite example. Components, native timing and internal route typing therefore need independent justification rather than the label “physical” alone.

The three realization roles and the selection obstruction define what must be supplied.

Choose an independently nominated target and a proposed physical descriptor. If conditions assigned the same descriptor value nevertheless differ in that target, no exact bridge through the descriptor can recover it. This is the practical meaning of fibre sufficiency: a representation cannot explain a distinction it has already erased.

The source proposes calibrated tests with declared tolerance, measurement error and regularity assumptions. For continuous descriptors, nearby rather than identical values require a bound on how rapidly the proposed bridge can change. A discrepancy beyond the allowed combination of proximity and error rejects that specified bridge.

The relevant constants and target cannot be chosen after seeing a failure merely to preserve the fit. Likewise, repeatedly training a new decoder for each mismatch would not test one fixed claim. A successful prediction on held-out interventions is more informative when the representation and calibration were frozen in advance.

For report distributions, compare the distributions rather than one coincident output. Physical instrument error, sampling uncertainty and model approximation have different origins and should remain identifiable in the analysis.

Chapter 23.2 gives the exact calibrated formulation.

A2 identifies phenomenal relational organization with the complete endogenous predictive object. To test that identification, the phenomenal comparison cannot simply be defined as a copy of the computed object. That construction would establish mathematical realizability of the assignment, not independent agreement with experience.

The monograph suggests nominated discrimination relations, reported similarities, temporal succession judgments or selected contrasts. These measurements have their own report channels and may not exhaust phenomenology. The test asks whether one fixed structural identification accounts for the nominated relations under held-out interventions.

The full predictive object retains more than a scalar contrast. Its richness preserves relevant labels and transitions, but also raises a risk: expanding the realization after every failure can make the account hard to falsify. A meaningful study specifies in advance which discrepancy counts against the application or law.

Calibrating finite record witnesses brings another requirement. A model can fit observations while its inferred witness remains unstable because the design is nearly singular. Report conditioning and physical recordability, not only fitting error. A formal observable is not automatically an accessible measurement.

Chapter 23.4 and finite witness calibration specify these obligations.

A1's exact nonzero-edge rule can change the assigned subject count under arbitrarily weak internal coupling. The worked merger begins with two qualifying persistence cores and, under a specified installed internal mechanism with compatible returns, assigns one combined core for every positive coupling strength. Finite-horizon physical transcript changes can remain very small.

This is a definite consequence, not a mathematical contradiction. It makes uncertainty and resolution important. A thresholded effective graph can be more stable, but it adds a convention rather than secretly reproducing the exact law. Studies should report whether uncertainty can change edges, core partitions or return qualification.

Native time matters too. Coarse sampling can hide a cycle or a qualification gap. Internal/external route typing can change individuation without removing physical communication. A researcher needs to justify the chosen organization and examine reasonable alternatives before treating the subject count as a discovery.

The same applies at the lower boundary. A truly certified one-bit return process can satisfy A1. Evidence of systematic over-admission in adequately specified non-toy applications would call for refinement. A complexity veto cannot be inserted without acknowledging a changed constitution.

The weak-coupling countermodel and A1's minimal-case discussion identify sharp assessment points.

The edition's proposed sequence starts with a limited experimental repertoire, an independently grounded report-feature family and a candidate physical model. Held-out conditions test descriptor sufficiency. Only then should the chosen decomposition and A2 correspondence be assessed. This order prevents a consciousness verdict from deciding how the data are first constructed.

Finite computational candidates can be checked more exhaustively under declared relabelings, protocol continuations, split and merge regimes and boundary perturbations. The two-bit example gives a transparent baseline, while a real organism or artificial system introduces hidden state, timing uncertainty, inaccessible variables and imperfect intervention.

The source suggests an open whole-brain Drosophila connectome plus an explicitly justified dynamical model as a possible first non-toy realization study. Its stated aim would be to investigate R* selection, decompositions, routes and approximate predictive structure. A wiring diagram alone neither determines native dynamics nor certifies that a fly is conscious.

The publication reports no new neural-data validation and no completed study of that proposed realization. Its available result is the finite conditional constitution, with mathematical checks and countermodels. The next evidential step is an independently constrained application that preserves failures and uncertainty, rather than a new label for an existing signal.

Chapter 23.6 states the proposed sequence; Chapter 24.6 lists what remains outside the theorem.