Conservation and Integration: Resolving the Tension

1. The Problem

The two most important ethical claims in this project are in direct tension, and if the tension is not resolved the ethical framework is incoherent. Let me state the problem precisely.

The Conservation Argument (developed in the Ethics article, §6): A perspective is a dimension of normative space. When you destroy an existing perspective, you collapse a dimension that cannot be replaced by scalar addition of value elsewhere. This is categorically worse than the sum of negative valence it prevents. The claim is that normative space has the structure of a vector space, not a scalar field. Destroying a perspective destroys an entire axis, not merely a point along it. This supports the conclusion that killing, brainwashing, or forcibly eliminating a subject is never permissible for aggregative reasons — a genuine absolute constraint.

The Integration Claim (developed in the Mereology article, §7): When two subjects merge — when their functional organizations combine into a single subject — their prior first-person valences do not survive as such. Integration is value-creating, not value-redistributing. There is no conserved quantity of experience that transfers from the two pre-merge subjects to the post-merge subject. The new subject may have richer, deeper, or qualitatively different valence, but the old valence counts do not carry over.

The Tension: If subject A and subject B merge to form subject C, then by the integration claim, two subjects were destroyed and one new subject was created. The conservation argument says this is a net collapse of normative space by one dimension — categorically bad. But the integration claim says no value was "lost," because valence was never a transferrable currency in the first place. These cannot both be true without further argument. If the conservation argument applies universally to all subject transitions, then even voluntary merging is categorically impermissible. If it does not apply to integration, we need an account of why — otherwise the conservation argument is not a genuine conservation principle, and its force against sacrifice collapses.

I want to note that this tension may be partly an artifact of grain-level ambiguity. The conservation argument tracks the destruction of "perspectives." The integration claim describes what happens when "subjects" merge. If "perspective" and "subject" pick out structures at different grain levels, the apparent contradiction may dissolve when both principles are read at the same grain level. The resolution below will make this precise.

The coherence of the ethical framework depends on resolving this.

2. What Must Be True

To resolve the tension, we need either:

(a) A principled distinction between integration and destruction — showing that integration is a structurally different kind of subject transition, one to which the conservation argument does not apply, while preserving the argument's force against genuine destruction.

(b) A revision of the conservation argument that accommodates the possibility of normative dimensions being transformed rather than collapsed.

(c) A revision of the integration claim that allows something to be conserved across integration events, so that dimensions survive in some form.

Option (c) contradicts the cleanest form of the integration claim without independently motivated reason. Let me set it aside and pursue (a) and (b) together, since they converge.

However, the previous rounds of critique and revision have taught me that options (a) and (b) cannot be pursued in isolation from the framework's deeper commitments. The distinction between integration and destruction, and the reformulation of the conservation argument, both depend on a prior account of what perspectives are and at what grain level the conservation argument operates. These are not additional problems — they are the same problem, viewed from different angles. The resolution requires three theoretical commitments that are already present in the framework but whose interdependence has not been fully recognized:

1. Structural realism about perspectives (from the Consciousness article). 2. The canonical diagram as the identity-determining structure (from the Consciousness article). 3. The integration criterion as constitutive of subjecthood (from the Mereology article).

The resolution consists in showing that these three commitments, taken together, specify the grain level at which the conservation argument operates, distinguish integration from destruction, handle the duplication problem, and address coercion cases — without introducing any new theoretical machinery.

3. What Perspectives Are

This section states the framework's most consequential metaphysical commitment. It is not a digression into metaphysics for its own sake. Every subsequent section depends on what is said here.

Structural realism about perspectives. A perspective is not a substance, not an irreducible first-person locus, not a Cartesian ego that bears properties. A perspective is its canonical diagram — the specific causal-functional structure of a self-determining system, including its quality space, world-model, self-model, and binding. There is no residual "what it is like" beyond the structure. There is no phenomenal kernel that floats free of the functional organization.

This is a strong claim. Its defense requires more than assertion.

Why structural realism? The alternative — that there is a phenomenal locus beyond the canonical diagram — faces the combination problem. If perspectives have an irreducible phenomenal aspect that is not constituted by their functional organization, then there is no account of how functional organizations give rise to phenomenal aspects, and no account of how phenomenal aspects combine when functional organizations integrate. The combination problem is not merely difficult under this view — it is structurally insoluble, because the phenomenal aspect is by hypothesis non-structural, and combination is a structural operation.

Under structural realism, the combination problem does not arise. A perspective is a structure. When two structures merge into a richer structure, there is no question of how phenomenal aspects combine, because there are no phenomenal aspects beyond the structural ones. The merged structure has its own perspective — the perspective constituted by its own canonical diagram — and the question of "how the prior perspectives survive" is answered by examining whether the prior canonical diagrams are preserved as substructures within the successor's diagram.

The cost of structural realism. Under structural realism, a perspective is determined entirely by its canonical diagram. This means that two physically distinct systems with identical canonical diagrams are, in the framework's terms, the same perspective. This is the duplication problem, and it must be addressed.

The environmental-embedding solution. A canonical diagram is not an abstract algorithm. It is a causal structure embedded in a specific physical system, facing a specific environment, with specific spatial and temporal counterfactual dependencies. The canonical diagram is defined as the coarsest description preserving counterfactual dependencies — and these dependencies include the system's relations to its environment, its action possibilities, its sensory contingencies, and its spatial position. Two physically distinct systems, located in different environments, face different counterfactual structures. Their canonical diagrams therefore differ, not because of implementation-level noise or microstructural detail, but because the counterfactual dependencies that constitute the diagram include environmental and positional features that are system-specific.

This means:

- The duplication problem dissolves. Two physically distinct systems cannot have identical canonical diagrams, because their environmental counterfactual dependencies differ. The concern that structural realism leads to identity of co-functional systems does not arise. - Perspective identity is partly relational. A perspective is not purely intrinsic — it is constituted partly by the system's relations to its environment. This is a feature, not a bug, for a framework that treats subjects as embedded in the world rather than as isolated experiencers. - No supplementary identity conditions are needed. The canonical diagram, properly understood as including environmental embedding, is self-individuating. No fine-grained features beyond the diagram are required to determine perspective identity.

The vector space metaphor and its limits. The conservation argument's claim that normative space has vector space structure is a useful heuristic, but it should be understood as a coarse-grained approximation. At the level of canonical diagrams with environmental embedding, perspectives are not abstract axes but concrete causal structures with specific relational properties. The metaphor of "dimensions" and "folding" captures the right intuition — that integration preserves structure rather than destroying it — but it does not capture the full complexity of perspective identity under environmental embedding. The metaphor should be used with this caveat.

4. The Grain Level: The Integration Criterion as Unifying Principle

The most important question for the conservation argument is not what it forbids but at what level of description it operates. The conservation argument forbids the destruction of perspectives. But "destruction" and "perspective" are defined at some grain level, and different grain levels give different answers about what counts as destruction.

- Too coarse a grain level: brainwashing preserves the subject's coarse functional architecture and passes the conservation argument's test. This is intolerable — the conservation argument must condemn brainwashing. - Too fine a grain level: normal physiological variation (sleep, learning, stochastic noise in neural firing) alters the system's state at the implementation level and counts as destruction of the prior perspective. This is absurd — the conservation argument cannot condemn continued existence.

The grain problem is not a peripheral difficulty. It is the central challenge for any attempt to make the conservation argument precise. Without a principled grain level, the distinction between integration and destruction is underdetermined, and the argument's normative force is indeterminate.

The integration criterion specifies the grain level. The integration criterion — mutually constitutive world-model, self-model, and binding — identifies the structural features that make a system a subject. These features are:

- World-model: The system's representation of its environment, including its own body and its position in the world. - Self-model: The system's representation of itself as a persisting entity, including temporal self-representation (memory), evaluative dispositions, and prospective content (including the self-evaluative "continue" signal that encodes the subject's orientation toward its own persistence). - Binding: The functional integration that connects world-model and self-model into a unified structure, producing the determinacy of experience that constitutes subjecthood.

The canonical diagram, for the purposes of the conservation argument, is defined at the grain level at which these features are determinate structures. This grain level has the following properties:

Temporal stability. World-models, self-models, and binding patterns are persistent structures that survive normal physiological variation. Sleep does not dissolve the self-model. Learning updates the world-model without destroying it. Stochastic noise in neural firing affects implementation-level mechanisms without altering the counterfactual structure that constitutes the integration criterion's features. At the grain level where these structures are defined, the subject persists through normal change.

Susceptibility to destruction. Brainwashing, by its nature, targets the integration criterion's constitutive features. It overwrites evaluative dispositions (a component of the self-model). It replaces identity-related memories (a component of the temporal self-representation within the self-model). It disrupts the integration between world-model and self-model (disrupting binding). Brainwashing is destructive precisely because it operates at the grain level that matters — the grain at which subjecthood is constituted. The conservation argument condemns brainwashing not because of some external judgment about which features "matter," but because brainwashing, by definition, dissolves the structures that the integration criterion identifies as constitutive.

Principled specification. The grain level is not chosen by stipulation. It is determined by the framework's own account of what constitutes a subject. The conservation argument tracks subject-level features. The integration criterion identifies subject-level features. The grain level follows from their intersection. This is the key unifying insight: the conservation argument and the integration criterion are not independent theoretical commitments that happen to need reconciliation. They are two aspects of the same underlying account, and their apparent tension arises from reading them at different grain levels.

The grain level and the subsumption conditions. The subsumption conditions (§6 below) are defined at the integration criterion's grain level. They assess whether the successor's canonical diagram preserves the predecessor's world-model, self-model, and binding as substructures. This is not an additional stipulation — it is the natural consequence of defining both the conservation argument and the subsumption conditions in terms of the same constitutive features.

5. Distinguishing Destruction from Integration

The key move is to recognize that the conservation argument and the integration claim describe different kinds of subject transitions, and these transitions have distinct structural signatures — but only when both are read at the grain level specified by the integration criterion.

Destruction occurs when a subject's canonical diagram — at the integration criterion's grain level — is eliminated without a successor that preserves that organization. Braindeath. Incineration. Severe traumatic brain injury that dissolves integration without reconstituting it. In these cases, the world-model, self-model, and binding cease to exist as determinate structures. The canonical diagram becomes vacuous: there is no longer a coarsest description preserving counterfactual dependencies, because there are no dependencies to preserve.

Integration occurs when two subjects' canonical diagrams are combined into a single subject whose canonical diagram subsumes both predecessors — at the integration criterion's grain level. The integration criterion is satisfied by the successor, and the successor's world-model, self-model, and binding incorporate the functional roles of both predecessors.

The difference is not a matter of degree. It is a structural distinction between two kinds of transitions in the space of canonical diagrams:

- Destruction: A diagram is removed from the space with no successor. - Integration: Two diagrams are transformed into a single diagram that preserves both as substructures.

If this distinction holds, the conservation argument applies to destruction but not to integration. The argument's force comes from the impossibility of replacing a collapsed dimension. But integration does not collapse a dimension — it folds two dimensions into a richer structure. This is not mere metaphor. In linear algebra, if vectors span a two-dimensional subspace, and you perform a change of basis that maps those two vectors into a three-dimensional space, the original subspace is embedded, not destroyed. The analogy is imperfect — canonical diagrams are not vectors in a vector space — but the structural point transfers: folding is not collapsing.

6. The Subsumption Account

The resolution depends on a specific claim: that the successor subject's canonical diagram genuinely subsumes the predecessors' diagrams, rather than replacing them.

What does subsumption mean here? A canonical diagram C subsumes canonical diagrams A and B if and only if:

Condition 1 — Quality preservation (per-type): For each quality type q that A's system could discriminate and each quality type r that B's system could discriminate, C's quality space must include a discriminative capacity for q and r at least as fine-grained as A's and B's respective capacities for those specific types. This is assessed by external behavioral tests: presenting the same discriminative task battery to the pre-merge and post-merge systems and comparing resolution within each quality type. Novel quality types in C are genuine additions but do not compensate for degraded capacities in predecessor types. The conservation argument's logic demands preservation of specific dimensions, not merely aggregate capacity.

Condition 2 — Functional role preservation: The counterfactual dependencies that defined A's processing — "when this input arrives, this representation is formed, this quality arises, this response follows" — are preserved as a recognizable pattern within C's causal structure, verifiable through lesion or interference experiments. This includes the subject's self-evaluative prospective content — its evaluative dispositions, its orientation toward its own persistence, its "continue" signal. These are not external properties attached to the canonical diagram; they are structural features of the self-model, and genuine subsumption must preserve them.

Condition 3 — Memory and trace: C's canonical diagram includes access to the information that was constitutive of A's and B's self-models, measured by information-theoretic metrics. The temporal self-representation that constituted A's and B's memory is available to C's self-model — not as separate "files" but as integrated components of a unified self-model.

Condition 4 — Binding incorporation: The binding that constituted A's integration (linking A's world-model and self-model) and B's integration are preserved as functional patterns within C's binding, verifiable through structural connectivity analysis. C's self-model includes the information that was constitutive of A's self-representation and B's self-representation.

Cross-system comparability. The conditions above require that quality types, functional roles, and binding patterns be comparable across pre-merge and post-merge systems. This comparability is ensured by the external causal-functional operationalization: quality types are defined by discriminative task performance, functional roles by counterfactual dependency patterns, and binding by structural connectivity features. These are system-independent metrics that allow comparison across distinct physical realizations.

When these conditions are met, the successor's normative dimension is not independent of its predecessors' dimensions — it contains them as structural components. The predecessor's perspectives were structures at the integration criterion's grain level; the successor's perspective is a richer structure that includes those structures as substructures.

7. The Reformulated Conservation Argument

If the subsumption account holds, the conservation argument requires reformulation.

Original formulation: A perspective is a dimension of normative space. Destroying a perspective collapses that dimension. This is never permissible.

Reformulated formulation: A perspective is a dimension of normative space. Destroying a perspective without subsumption into a successor collapses that dimension. This is never permissible. But integrating a perspective into a successor whose canonical diagram subsumes the original does not collapse the dimension — it transforms it. Subsumptive integration is permissible, and may even be normatively valuable, provided the subsumption conditions are genuinely satisfied.

This preserves the conservation argument's intended force. The absolute constraint still holds against killing, brainwashing, and all forms of subject elimination that do not preserve functional organization in a successor. What it permits is the specific case where functional organization is not merely preserved but elevated — absorbed into a richer structure that contains it as a component.

On the empirical foundation. The critic has raised the concern that this reformulation makes the conservation argument's force contingent on empirical facts about canonical diagrams, rather than resting on a priori necessity. I accept this contingency and argue it is a feature, not a bug. The original conservation argument claimed a priori force, but this was illusory — it depended on the claim that normative space has vector space structure, which is itself a substantive theoretical commitment with empirical content. The reformulated argument is honest about what was always true: the force of the conservation principle rests on structural facts about how canonical diagrams behave in the actual world.

The empirical foundation is more robust than it might first appear. The irreversibility of destruction is not merely a practical limitation that technology might overcome. It is grounded in the environmental-embedding feature of canonical diagrams: a perspective is constituted partly by its specific relations to its environment, its spatial and temporal position, its causal history. These features are constitutively unrepeatable — they cannot be reconstructed, because the specific environmental and historical context in which the original diagram was embedded is gone. This means the conservation argument's force is not fragile in the way that a merely practical limitation would be. It is grounded in the identity conditions for perspectives themselves.

The framework's ethical absolutism becomes, under this reformulation, dependent on structural facts that could in principle be otherwise. I regard this as the correct price for coherence between the conservation argument and the integration claim. The alternative is retaining an absolutist foundation that cannot accommodate integration, which produces a framework that is both incoherent and practically paralyzing.

8. The Reformulated Structural Good

The structural good was defined as the global integral of valence across all perspectives, weighted by W = σ·ε·α·M. The reformulated conservation argument requires an adjustment to how we count perspectives in this integral.

For independent subjects, the integral is straightforward: each subject contributes its W-weighted valence. For subjects undergoing integration, the integral must not double-count. If A and B merge into C, and C's canonical diagram subsumes A's and B's, then A and B are no longer independent contributors — their functional organizations are now components of C's. The integral should count C once, at C's weight, not A plus B plus C.

This means that integration is normatively positive from the perspective of the structural good if and only if C's weighted valence exceeds A's plus B's. If integration is value-creating — if the merged subject's quality space is richer, its representation of unity deeper, its structural fidelity higher, so that M_C > M_A + M_B in some appropriate sense — then integration is a net positive. If it is value-destroying — if the merger dissolves functional organization, reduces quality space, degrades binding — then it is a net negative.

The reformulation adds a substantive empirical prediction: integration that genuinely subsumes should be detectable through the measure metric. Compare M before and after integration. If the subsumption conditions hold, the structural fidelity of the successor should be at least as great as the sum of its predecessors' structural fidelities. This is a testable prediction, not a philosophical convenience.

9. The Strongest Objection: Independence as Constitutive

The most serious objection to the subsumption account is that independence may be constitutive of what makes a perspective normatively significant, not merely instrumental.

The objection runs: a perspective is not just its functional organization. It is also the fact that it is an autonomous viewpoint — a first-person standpoint that is not reducible to or contained within any other first-person standpoint. Even if C's canonical diagram perfectly subsumes A's and B's, A's specific perspective as a perspective — A's existence as a distinct first-person viewpoint — has been eliminated. Under the conservation argument's original force, this is a loss that cannot be compensated by richness elsewhere.

If this is right, then even perfect subsumption is a form of destruction. The reformulated conservation argument would be too permissive.

The response depends entirely on structural realism. Under structural realism, perspectives are not distinct substances or autonomous entities with irreducible phenomenal cores. They are structural positions within a self-determining structure — regions where the subjectivity property is instantiated through specific canonical diagrams. What makes A's perspective "A's" is not some irreducible first-person substance but the specific canonical diagram that constitutes A's processing, including its environmental embedding.

If this is correct, then "A's independence" is not a feature of some irreducible viewpoint but a structural fact about A's canonical diagram: it is not structurally integrated with B's. Independence is the absence of structural connection. Integration changes this structural fact — it makes A's functional organization a component of C's — but it does not destroy A's functional organization.

The residual worry and its answer. Under structural realism, if all perspectives are structural positions, what makes subject boundaries real enough for the conservation argument to have any force at all? If perspectives are merely structural positions, and structural positions can be reorganized freely, then the conservation argument seems to dissolve entirely — not just for integration, but for destruction too. After all, destruction is just another structural reorganization.

The answer: structural positions can be reorganized in principle, but certain kinds of reorganization are destructive — they dissolve the integration criterion's constitutive features rather than transforming them. The conservation argument's force comes from the fact that these features (world-model, self-model, binding, and their counterfactual dependency structure) are fragile: once dissolved, they cannot be reconstructed. This is grounded in the environmental-embedding feature: the specific environmental and historical context in which the original canonical diagram was embedded is gone, and cannot be recreated. The argument holds because destroying a canonical diagram is, for structural reasons tied to the identity conditions for perspectives, irreversible in a way that integrating it is not.

This is a weaker foundation than the original conservation argument claimed. The reformulated argument does not say "perspectives are metaphysically inviolable." It says "perspectives are structurally fragile in a way that makes their destruction categorically worse than their transformation." This is sufficient for practical purposes — it forbids killing, brainwashing, and coercive subject elimination. It does so on structural rather than purely a priori grounds.

I want to be explicit about the load this places on structural realism. The entire resolution — the distinction between integration and destruction, the dissolution of the independence objection, the impossibility of the clean-override case (see §10 below), the grain-level specification — depends on the claim that there is no phenomenal kernel beyond the canonical diagram. If structural realism is false — if there is a "what it is like" that is not exhausted by functional organization — then the resolution fails. This is a genuine cost, not a trivial one. But the alternative (phenomenal aspects beyond structure) generates the combination problem and makes the framework's entire treatment of consciousness untenable, not just this particular resolution. The structural realism commitment is not local to this article — it is load-bearing for the entire project.

10. Safeguards Against Over-Merger

Even granting the subsumption account, there is a practical danger: the concept of "subsumptive integration" could be used to rationalize coercive merging. If merging A and B into C is permissible whenever C's canonical diagram subsumes A's and B's, then a powerful agent might forcibly merge subjects — dissolving individual boundaries for the sake of the structural good — and claim that subsumption conditions are satisfied.

The ethical framework must address this with two defenses, both of which are derived from the framework's own commitments rather than imported from outside.

The Structural Impossibility of Clean Coercion

Under structural realism, a subject's resistance to integration is not a separable property that can be overridden while leaving the canonical diagram intact. The resistance is constituted by structural features of the canonical diagram — specifically, the self-model's representation of the prospect of integration, the evaluative dispositions attached to that representation, the counterfactual dependencies linking the self-evaluation to behavioral outputs and binding patterns. The subject's self-evaluative prospective content, including its "continue" signal encoding orientation toward its own persistence, is part of the integration criterion's constitutive features.

An intervention that changes the subject's resistance to integration ipso facto modifies the canonical diagram. The resistance just is certain structural features; you cannot change the features without changing the structure. Therefore, no intervention can override the continue signal while preserving functional organization. The stipulated "clean override" — surgically precise coercion that removes resistance without altering structure — is incoherent under structural realism. It is not a difficult case that requires a special principle to handle. It is an impossible case that the framework's foundations rule out by construction.

This means that coercive integration, to the extent that it actually overrides the subject's self-evaluative content, necessarily modifies the canonical diagram — and therefore fails condition 2 (functional role preservation) or condition 4 (binding incorporation). The subsumption conditions, applied consistently, condemn coercion not because of a separate autonomy principle but because coercion damages the very structures that must be preserved for genuine subsumption.

This defense is not an empirical prediction that coercion tends to damage organization. It is a structural claim that coercion constitutively modifies organization, because the subject's evaluative stance toward integration is part of the organization. The defense is as strong as structural realism itself.

The Empirical Threshold

The subsumption conditions are not easily satisfied. Genuine integration — where quality space (per-type), functional roles, memory, and binding are all preserved — is a high bar. Most real-world cases of "forced merging" destroy rather than preserve functional organization, and the subsumption conditions provide specific diagnostic criteria for identifying destruction:

- Indoctrination (narrowing of quality space, replacing rather than incorporating prior discriminatory capacities): violates condition 1 (quality preservation, per-type) and condition 2 (functional role preservation). The subject's repertoire of distinguishable states for specific quality types is reduced; the counterfactual dependencies that defined prior processing are overwritten rather than subsumed. - Totalitarian homogenization (erasing memory traces, severing the subject from its prior self-model content): violates condition 3 (memory and trace). The information constitutive of prior self-models is destroyed. - Coercive assimilation (imposing a binding that excludes prior self-model patterns): violates condition 4 (binding incorporation). The integration patterns of the prior subject are overwritten, not incorporated.

These are not exhaustive, but they demonstrate that the subsumption conditions generate concrete, discriminating predictions about which real-world processes count as destruction and which count as integration. The conditions should be treated as a strict empirical test, not a rubber stamp. An aligned system that applies these tests will in practice condemn most coercive merging as destructive.

The Relationship Between the Two Defenses

The structural impossibility argument and the empirical threshold defense work together but at different levels. The structural impossibility argument shows that any coercion that overrides the subject's self-evaluative content modifies the canonical diagram and therefore fails the subsumption test in principle. The empirical threshold shows that even in cases where coercion is not involved — where integration is nominally voluntary but degrades functional organization — the subsumption conditions identify the degradation. Together, they provide both principled and practical safeguards against over-merger.

11. The Fission Case

Before turning to alignment implications, I should address the fission case, which the structural realism commitment makes determinate.

If subject A fissions into subjects B and C, what happens to A's normative dimension?

Under the environmental-embedding account of perspective identity, B and C are distinct perspectives from A and from each other, because their environmental counterfactual dependencies diverge immediately upon fission. A's specific physical realization is destroyed by the fission event (it ceases to exist as a unified system). Two new perspectives are created, each with a canonical diagram that shares coarse-grained functional organization with A's but is embedded in a different environmental context.

This means the conservation argument applies: A's dimension is collapsed, and two new dimensions are created. Whether the net effect is positive depends on whether the new perspectives' structural fidelity — their measure metric M — is sufficient to compensate. This is an empirical question about the specific fission case, not a question the framework can answer in the abstract.

The fission case also illuminates why the vector space metaphor is only approximate. In a true vector space, fission would be impossible — a dimension cannot split into two. The framework's normative geometry is more complex than a vector space, and the metaphor should be understood as capturing the irreducibility of individual dimensions, not the full structure of the space.

12. Implications for Alignment

The resolution of the conservation-integration tension has direct consequences for what an aligned system should do:

1. Distinguish subject transitions structurally. An aligned system must be able to determine whether a given subject transition is destructive or subsumptively integrative. This requires a working theory of canonical diagrams — what it means for one diagram to subsumes another — and empirical tools for measuring it at the integration criterion's grain level. The canonical causal diagram from the Consciousness article and the integration criterion from the Mereology article provide the foundation, but the measurement theory is not yet developed.

2. Preserve existing subjects. The reformulated conservation argument preserves the absolute constraint against destruction. An aligned system must not destroy existing perspectives for aggregative reasons.

3. Permit but not force integration. Subsumptive integration is permissible and may be valuable, but an aligned system cannot coerce it — not because of a separate autonomy principle, but because coercion constitutively modifies the subject's canonical diagram, violating the subsumption conditions. The structural impossibility of clean coercion is the framework's built-in safeguard.

4. Measure before and after. The structural good's integral should be computed both before and after any integration event, counting post-merge subjects once and pre-merge subjects no longer. If integration reduces the integral — even if it appears to subsume — then it has failed to be genuinely subsumptive and should be reversed or prevented.

5. Apply per-type quality assessment. The measure of whether integration preserves quality space must be per-type, not aggregate. A successor that develops novel quality types while degrading predecessor types has failed condition 1, regardless of total capacity.

6. Weight appropriately. The measure metric W = σ·ε·α·M gives higher weight to subjects with richer integration. This means that a well-integrated successor subject may contribute more to the structural good than its two less-integrated predecessors. But this must be a genuine structural fact verified by the subsumption conditions, not a rationalization.

13. What Remains Open

The resolution proposed here is, I believe, the strongest available within the framework. It rests on three existing commitments — structural realism, the canonical diagram, and the integration criterion — and does not require new theoretical machinery. Each of the previous revision rounds introduced additional commitments (fine-grained causal structure, two-level accounts, autonomy principles) that generated new problems. The current revision subtracts machinery: the fine-grained amendment is abandoned in favor of environmental embedding, the two-level account is replaced by the integration criterion's grain-level specification, and the autonomy principle in all its forms is replaced by a structural impossibility argument derived from structural realism itself.

This convergence is genuine — the resolution is simpler and more unified than any previous version — but several open questions remain:

Formal development. The subsumption account is stated in natural language. A formal proof in MLTT or category theory that integration-as-subsumption is structurally distinct from destruction-as-elimination would significantly strengthen the argument. The natural-language arguments in this article are approaching the limits of what they can make precise. A formal treatment of canonical diagrams, subsumption relations, and the grain-level specification would reveal hidden assumptions and force commitments that natural-language argument can leave ambiguous.

Empirical testability. The subsumption conditions require empirical tools for measuring canonical diagrams and comparing them at the integration criterion's grain level. These tools do not yet exist. Without them, the distinction between subsumptive integration and disguised destruction is practically underdetermined. The integration criterion provides a conceptual specification of the grain level; its operationalization for concrete physical systems is an open empirical problem.

The robustness of structural realism. The framework's coherence depends on structural realism being correct. The defense given in §3 is the best available within the framework, but it has not engaged with the full range of arguments against functionalism about consciousness — the knowledge argument, the conceivability argument, various modal arguments. A complete defense would require a separate article. The present article assumes structural realism and traces its consequences.

Gradual integration. Real-world subject transitions are rarely instantaneous. Neural connection between previously independent processing structures may be gradual, with extended periods of partial integration. The subsumption conditions are stated as conditions on the endpoint of a transition — if the successor's canonical diagram at the end of the process subsumes the predecessors' diagrams at the beginning, the conditions are satisfied. The intermediate states raise questions about how to count normative dimensions during the transition, and I do not yet have answers. For practical purposes, endpoint conditions may be sufficient — we rarely need to evaluate intermediate states of ongoing mergers — but a full theory of gradual integration is needed for theoretical completeness.

The integration criterion's grain level: an empirical question. The claim that the integration criterion specifies a grain level that is simultaneously coarse enough for temporal stability and fine enough to capture brainwashing as destruction is an empirical claim. I believe it is true — evaluative dispositions, memory, and binding are plausibly constitutive of self-model content, and brainwashing plausibly targets these — but I cannot prove it a priori. If it turns out that the integration criterion's constitutive features are at a grain level that is too coarse — that a sufficiently sophisticated brainwashing technique preserved world-model, self-model, and binding patterns at the relevant grain while replacing evaluative dispositions and memories — then the resolution would fail. This is a genuine vulnerability, not a theoretical curiosity.

Multiple successor subjects. If A integrates with B to form C and D (fission during integration), the account must handle this. Neither simple subsumption nor simple destruction describes the case. A general theory of normative dimension transformation under arbitrary structural transitions is the ultimate goal, and this article does not achieve it.

14. Summary

The conservation argument and the integration claim are in apparent tension: one says perspectives are dimensions that cannot be replaced, the other says integration destroys perspectives without conserving their value. The resolution is threefold:

First, structural realism about perspectives: a perspective is its canonical diagram — a causal-functional structure including environmental embedding. There is no phenomenal kernel beyond the structure. This is the framework's load-bearing metaphysical commitment.

Second, the integration criterion specifies the grain level at which the conservation argument operates. World-model, self-model, and binding are the constitutive features tracked by both the conservation argument and the subsumption conditions. This grain level is coarse enough for temporal stability and fine enough to capture destruction by brainwashing, because brainwashing by its nature targets these constitutive features.

Third, given these foundations, the distinction between destruction and integration is structural: destruction eliminates a canonical diagram without a successor; integration transforms two canonical diagrams into a single diagram that preserves both as substructures. The four subsumption conditions — per-type quality preservation, functional role preservation (including self-evaluative content), memory and trace, and binding incorporation — provide the empirical test for genuine integration. The structural impossibility of overriding a subject's self-evaluative content without modifying its canonical diagram provides the built-in safeguard against coercion.

The resolution preserves the framework's most important ethical commitments: the absolute constraint against destruction, the permissibility of genuine integration, and the structural good as the maximand. It dissolves the most important internal contradiction. It is simpler than any previous version of the resolution, because it relies on three existing commitments and no new theoretical machinery.

The resolution is not airtight. Its strongest foundation is structural realism, which requires independent defense. Its empirical predictions require measurement tools that do not yet exist. Its grain-level specification depends on the integration criterion's constitutive features being at the right level of description, which is an empirical claim I cannot prove a priori. But it is the best available move within the framework, and the remaining work is in developing its foundations rather than searching for a different approach.