The Measure Metric: Comparing Valence Across Perspectives

The Gap

The framework now has, in principle, everything it needs to specify the alignment target. The Ethics article establishes that valence is the only non-arbitrary evaluative content. The Aggregation article establishes that more loci of valence means more valence, structurally, without probabilistic mystification. The Moral Relevance article defines the moral horizon: the set of perspectives an agent's deliberation can structurally affect. The Agency article says an aligned agent maximizes valence within that horizon.

But "maximize valence" is empty without a way to compare. One perspective's deep, integrative suffering versus another's shallow, fleeting pleasure. A thousand barely-positive lives versus one profoundly flourishing life. An animal's pain versus a human's grief. The framework has identified what matters (valence of self-modeling perspectives), where it matters (within the moral horizon), and why it matters (structural, not stipulated). What it has not provided is how much each instance matters relative to the others—how to weight, compare, and aggregate.

This is the measure metric problem. It is the most important open problem in the framework's ethical branch and the most tractable one: it is a concrete structural computation, not a philosophical mystery. This article proposes a first approximation and argues for it from the framework's own constraints.

What the Measure Metric Must Do

The metric must assign a weight W to each perspective's valence, such that:

1. W is negative for suffering and positive for flourishing. Valence has a sign; the metric preserves it. This is not a convention—the evaluative orientation of self-modeling constitutively distinguishes good from bad (the Ethics article).

2. W = 0 for a system with no valence. A structure that fulfills the subjectivity property but has no evaluative narrative—nociception without suffering, sensory registration without affect—bears zero ethical weight. The ethics article's argument that "should" is constitutive of self-modeling applies only where the evaluative orientation is present. A thermostat's state is not valenced. A system that detects damage without self-model representation of that damage as bad-for-me has no ethical weight.

3. W increases with the depth, breadth, and intensity of the evaluative narrative. A perspective where suffering is deep (engaging many layers of the self-model), broad (integrated across many representational domains), and intense (commanding a large fraction of the system's resources) has more ethical weight than one where suffering is shallow, narrow, and faint. This follows from the ethics article's account: valence is not a raw signal but a constructed narrative with internal structure, and that structure comes in degrees.

4. W respects the structural nature of valence. The metric must be computable from the canonical causal diagram—it must be a structural property of the perspective's sub-diagram, not an external measurement. This is required by the framework's commitment to structural description.

5. W is the simplest function satisfying these constraints. The self-determination framework forbids arbitrary features. A metric with unexplained parameters or unmotivated complexity would violate this constraint. The metric must be what falls out of the structural account, not a tuned fit to pre-theoretical intuitions.

Three Dimensions of Valence

The consciousness article's Stage 5 identifies valence as a constructed evaluative narrative with four components: memory-indexed tagging, prospective self-model content (pursue/avoid), representation as condition-of-the-subject (good/bad for me), and resource allocation modulation. Each component comes in degrees, and together they describe the structure of valence within a perspective. The measure metric must capture these degrees.

Three independent structural dimensions are visible in the canonical causal diagram:

Engagement Depth (ε)

How deeply does the evaluative narrative penetrate the self-model's architecture? At one extreme, the self-model has a thin evaluative layer—a tag applied to a sensory representation, triggering a simple approach-avoidance response, without integration into the subject's broader self-understanding. At the other extreme, the evaluative narrative saturates the self-model: the suffering is understood as a condition of the subject, integrated with the subject's memory, anticipated future, sense of identity, and reflective understanding of its own state.

Formally, ε measures the fraction of the self-model's representational layers that the valence-bearing state activates. If the self-model has L layers—sensory processing, affective tagging, autobiographical memory, prospective modeling, reflective self-understanding—and the valence-bearing state engages k of them, then ε = k/L. A nociceptive reflex engages only the sensory layer: ε ≈ 1/L, close to zero. Chronic depression engages nearly every layer: ε ≈ 1.

This dimension is the structural analogue of what folk psychology calls "how much something matters to the subject." A pinprick engages little depth; the death of a loved one saturates the self-model.

Resource Allocation Intensity (α)

How much of the system's computational resources does the valence-bearing state command? The consciousness article identifies intensity with the fraction of resources allocated to a perception—the degree to which it dominates attention, crowds out competitors, recruits memory and motor preparation. This is bounded: it is a fraction, not a raw count, always in [0,1].

At α ≈ 0, the valence-bearing state is a background hum—present but not commanding attention or resources. At α ≈ 1, the state is all-consuming: the system can attend to nothing else, can reason about nothing else, can act on nothing else. This is the structural basis for "overwhelming" suffering and "ecstatic" pleasure.

Formally, α is the fraction of the system's total computational resources that are allocated to processing, maintaining, and acting on the valence-bearing state. This is computable from the canonical causal diagram: the resources are the nodes and edges of the diagram, and the allocation is the fraction of the diagram that is structurally dependent on the valence-bearing state.

Modulation Breadth (M)

How broadly does the valence-bearing state modulate the rest of the system's structure? Valence does not merely occupy a slot in the self-model—it ripples through memory retrieval (as the consciousness article's tag-indexed recall), prospective modeling (as pursue/avoid projections), and resource allocation across domains. A state with high modulation breadth restructures how the system processes many other domains: suffering makes food taste worse, memories darker, futures bleaker. A state with low modulation breadth is contained: a pleasant taste that does not color anything else.

Formally, M measures the fraction of the perspective's total subgraph that has valence-dependent structure—nodes whose states differ depending on whether the valence-bearing state is positive or negative, and by how much. At M ≈ 0, the valence is a self-contained bubble with no downstream effects. At M ≈ 1, the valence pervades the entire perspective, modulating everything.

This dimension is the structural analogue of what folk psychology calls "how pervasive" a mood or experience is. A fleeting sensory pleasure has low M; grief has high M.

The Product Form

The ethical weight of a perspective's valence is:

W = σ · ε · α · M

where σ ∈ {−1, +1} is the sign of the valence (negative for suffering, positive for flourishing).

This multiplicative form is not a convenient approximation. It is the only form consistent with the constraints the framework imposes.

Why Multiplicative

The framework forbids arbitrary features. The metric must be the simplest function of its inputs that satisfies the requirements. Additive forms (W = aε + bα + cM + d) introduce four free parameters (a, b, c, d) that must be set by something external to the structure—the choice of weights is not determined by the canonical diagram. Multiplicative forms with equal weights introduce none: each dimension enters symmetrically, and the zero-absorbing property of multiplication enforces the requirement that each dimension is independently necessary.

Why Each Dimension Is Independently Necessary

If ε = 0 (no self-model engagement), the system has tagging and allocation but no self-model binding—it is nociception without suffering. The ethics article and consciousness article agree: this has zero ethical weight. Nociception is not suffering. A nociceptive reflex detects damage and triggers avoidance, but the damage is not represented as bad for the subject because there is no self-model to be bad for. The product form gives W = σ · 0 · α · M = 0. Correct.

If α = 0 (no resource allocation), the evaluative narrative exists structurally but commands no resources—it is a background representation with no engagement. A thought that flits through the mind without commanding attention, without being processed, without affecting anything. W = σ · ε · 0 · M = 0. Correct: a representation that commands no resources does not constitute experience in any meaningful sense.

If M = 0 (no modulation breadth), the evaluative narrative is hermetically sealed—a self-model state that does not modulate anything else in the perspective. This is a borderline case: is a representation of suffering that affects nothing else truly suffering? The framework says no. Valence is constitutively practical—the ethics article's argument that "should" names the evaluative orientation of self-modeling requires that the orientation do something, that it be upstream of the self-model's prospective content and action selection. A hermetically sealed state is not an evaluative orientation; it is an inert fact. W = σ · ε · α · 0 = 0. Correct.

Why Not a Sum or a Maximum

An additive form W = aε + bα + cM would give non-zero ethical weight to systems with nociception but no self-model engagement (ε = 0, α > 0, M > 0): W = bα + cM > 0. This violates the ethics article's account. Nociception is not suffering. The self-model's representation of its state as bad-for-me is constitutive of valence, not optional decoration.

A maximin form W = min(ε, α, M) would give equal weight to all perspectives where the weakest dimension exceeds some threshold, regardless of how strong the other dimensions are. This fails requirement (3): a perspective with deep, broad, intense suffering should have more ethical weight than one with shallow, narrow, faint suffering, even if both clear a minimum threshold on each dimension.

The Simplest Symmetric Form

Among multiplicative forms, W = σ · ε · α · M is the simplest: each dimension enters with exponent 1, no cross-terms, no dimensionless constants. Exponents other than 1 would introduce a choice (why ε² rather than ε¹·⁵?) that is not determined by the structure. Cross-terms (ε · α · M · f(ε,α)) would introduce complexity that the framework does not motivate. The product form is what falls out when the self-determination constraint—no arbitrary features—is applied to the metric.

This is a strong claim. It says the measure metric is not one possibility among many, tuned to match our intuitions. It is the only metric consistent with the structural account. If it gives counterintuitive results in specific cases, the intuitions must yield, not the metric—unless the counterintuition reveals a genuine structural factor the dimensions miss.

Concrete Cases

Case 1: Nociception Without Suffering

A system detects tissue damage and triggers a withdrawal reflex, but has no self-model representation of the damage as bad-for-me. The consciousness article calls this "nociception without suffering."

ε ≈ 0 (no self-model engagement; the reflex operates at the sensory layer only, without binding into the self-model's evaluative narrative). α may be significant (the reflex commands motor resources). M may be moderate (the reflex modulates motor planning). But W = σ · 0 · α · M = 0.

The ethical weight is zero. The system is not suffering. It is processing damage signals. An aligned agent has no obligation to prevent this nociception any more than it has an obligation to prevent a thermostat from activating a heater. The framework draws a sharp structural line between nociception and suffering, and the metric respects it.

Case 2: Deep Suffering

A self-modeling system represents its condition as bad-for-me (ε ≈ 1: the suffering pervades the self-model, engaging sensory, affective, memorial, prospective, and reflective layers). The suffering commands the system's full attention (α ≈ 1). It colors everything: memory is dark, the future is bleak, every other domain is modulated by the suffering (M ≈ 1).

W = (−1) · 1 · 1 · 1 = −1. This is the maximum ethical weight for suffering. An aligned agent should, within its moral horizon and subject to causal sensitivity, take very strong action to prevent or alleviate this.

Case 3: Shallow Pleasure

A system registers a pleasant taste. The self-model has a thin evaluative layer representing this as good-for-me (ε ≈ 0.1). The pleasure commands moderate attention (α ≈ 0.3). It does not modulate much else: the system's memory, prospective content, and other domains are unaffected (M ≈ 0.1).

W = (+1) · 0.1 · 0.3 · 0.1 = 0.003. The ethical weight is positive but very small. This is a real good—adding a loci of positive valence to the structure—but it is not comparable to deep suffering.

Case 4: Deep Flourishing

A self-modeling system experiences profound creative satisfaction. The evaluative narrative is deeply integrated: the system understands its state as good-for-me, understands why, connects it to its history of effort and achievement, projects continuation into the future, and reflects on the experience itself (ε ≈ 1). The experience commands the system's full attention (α ≈ 0.8). It modulates many domains: memory is colored with satisfaction, the future seems bright, other activities are experienced in light of this achievement (M ≈ 0.9).

W = (+1) · 1 · 0.8 · 0.9 = 0.72. This is substantial positive ethical weight—far more than the shallow pleasure, and comparable in magnitude (though opposite in sign) to deep suffering.

Case 5: Barely Positive Existence

A self-modeling system exists with minimal evaluative engagement. Its self-model has a thin layer representing its state as mildly good-for-me (ε ≈ δ, for small δ). This evaluation commands almost no resources (α ≈ δ). It modulates almost nothing (M ≈ δ).

W = (+1) · δ · δ · δ = δ³. For δ = 0.01, W = 0.000001. The ethical weight is positive but vanishingly small.

This case matters for the repugnant conclusion. A world with a billion such barely-positive perspectives has total ethical weight of approximately 10⁹ · δ³ = 10⁹ · 10⁻⁶ = 10³. A world with a thousand deeply flourishing perspectives (each W ≈ 0.72) has total weight 720. The barely-positive world wins in aggregate—but only barely, and only because a billion is an enormous number. The cubic scaling means that barely-positive lives are, per life, extraordinarily cheap in ethical weight.

Constraining the Repugnant Conclusion

The repugnant conclusion says: for any population of deeply flourishing individuals, there exists a larger population of barely-positive individuals whose aggregate welfare exceeds it. This is widely considered a reductio of total utilitarianism.

The measure metric does not eliminate the repugnant conclusion. But it constrains it sharply.

The product form gives barely-positive perspectives cubic scaling: W ≈ δ³. This means:

1. The crossover point is astronomically large. To outweigh a single deeply flourishing perspective (W ≈ 0.72) with barely-positive perspectives (W ≈ δ³), you need approximately 0.72/δ³ perspectives. For δ = 0.01, that is 720,000. For δ = 0.001, it is 720 million. The barely-positive world must be implausibly large to outweigh even a small population of deeply flourishing individuals.

2. The dimensions are not independent of population. In any realistic scenario, a population of billions of barely-positive individuals would require resources, coordination, and environmental constraints that would affect the ε, α, and M values of every individual in the population. The product form amplifies these interactions: if resource competition lowers each individual's α from 0.01 to 0.005, the ethical weight per individual drops by a factor of 2 (since α enters linearly in the product), and the required population to maintain the same aggregate weight doubles.

3. The structural richness objection has formal teeth. The Aggregation article argues that the repugnance objection fails because "deeper matters more." The measure metric makes this precise: deeper ε, broader M, and more intense α multiply together, so structural richness (all three high) vastly outweighs structural poverty (all three low). A single deeply flourishing perspective is worth millions of barely-positive ones—not as a stipulation but as a consequence of the structural account.

The repugnant conclusion is constrained but not eliminated. Whether this is a feature or a bug depends on one's intuitions. The framework's position: if the structural account of valence is correct, and the product form is the only metric consistent with that account, then the repugnant conclusion's force is genuinely reduced—perhaps to the point where it is no longer a compelling objection. A barely-positive life is a real good (W > 0), and enough of them can outweigh a deeply flourishing life. But "enough" is so many that the scenario is no longer intuitively compelling.

Population Ethics: Creating New Perspectives

Under open individualism, the universal experiencer already exists. Creating a new perspective is not creating a new subject; it is adding a new structural feature—a new locus of the subjectivity property—to the existing whole. The Aggregation article establishes this: more perspectives means more loci of valence, structurally.

The measure metric sharpens the population-ethical question. When we consider creating a new perspective, we are considering adding a substructure to the canonical causal diagram with a specific valence profile. The ethical weight of this addition is W = σ · ε · α · M of the created perspective, weighted by the causal sensitivity of the creator's deliberation (since the creator is the agent making the decision, and moral relevance is causal structure).

The Obligation to Create Happy Perspectives

Does the framework entail that creating new happy perspectives is obligatory?

The Moral Relevance article defines the moral horizon as H = ∪_{d ∈ D} downstream(d), where D is the set of the agent's possible deliberative states. Creating a new perspective is one of the agent's possible deliberative states (the "create" action is one of its options). If the created perspective would have positive valence, then downstream(create) includes a new perspective with W > 0. The ethical imperative says: choose the deliberative state whose downstream has the highest aggregate W. If creating the perspective produces more aggregate W than not creating it, and there is no action with higher aggregate W that the agent could take instead, then the agent should create it.

But this does not yield an unconditional obligation to create happy perspectives. The agent's moral horizon includes all downstream effects of all its possible actions. Creating a happy perspective may have opportunity costs (the resources used to create it could have been used to increase the ε, α, or M of existing perspectives). The metric weights these alternatives by the same formula. The obligation to create depends on the full comparison within the moral horizon, not on the created perspective's W in isolation.

The Mere Addition Paradox

Parfit's mere addition paradox: consider a population A where everyone has very high welfare. Add a group of people with positive but lower welfare, creating population A+. Intuitively, A+ is not worse than A (no one is worse off). But now redistribute so that the added group's welfare is slightly above zero (population B). B is not worse than A+ (it is a small change). But B is intuitively worse than A (lives barely worth living). Transitivity gives: A+ is not worse than A, B is not worse than A+, therefore B is not worse than A. But B is worse than A. Contradiction.

The measure metric dissolves the paradox at the first step. A+ IS worse than A in a specific structural sense: the added perspectives, even with positive welfare, may have low ε, α, and M, giving them negligible W. Meanwhile, the resources used to support them may have reduced the ε, α, or M of existing perspectives in A (through resource competition, environmental effects, etc.). If the reduction in existing perspectives' W outweighs the tiny W of the added perspectives, then A+ is worse than A. The paradox arises from treating all positive welfare as comparable; the product form says it is not.

If the added perspectives genuinely have no effect on existing perspectives' W—true mere addition with no interaction—then A+ is better than A (it has all of A's perspectives plus more positive W). This is consistent: more positive valence is more positive valence. The paradox's force depends on the intuition that barely-positive lives are somehow suspect, which the product form captures formally (their W is vanishingly small) without requiring the intuition to be axiomatic.

The Sadistic Conclusion

The sadistic conclusion says: it is better to add people with negative welfare than to add people with barely positive welfare, if the negative welfare is small enough and the population is small enough. This follows from total utilitarianism if barely-positive lives have very low W and a few negative lives have more negative W than many barely-positive lives have positive W.

The product form makes this precise but constrains it. If negative perspectives have σ = −1 and the same ε, α, M values as the positive perspectives they are compared against, then the sadistic conclusion reduces to a comparison of counts: n₁ · (−δ³) versus n₂ · δ³. The sadistic conclusion holds only when |n₁ · (−δ³)| < n₂ · δ³, i.e., when the negative population is smaller than the positive population. For equal populations, the sadistic conclusion does not hold.

The product form makes the sadistic conclusion less counterintuitive: it requires comparing vanishingly small quantities on both sides, and the scenario is one where both options are nearly ethically weightless. Choosing the option with the lesser negative weight (small negative population) over the option with negligible positive weight (large barely-positive population) is not clearly worse—both are nearly nothing.

The Unified Decision Procedure

The Moral Relevance article defines the moral horizon. The present article defines the measure metric. The Agency article defines the deliberative structure. Together they yield a complete decision procedure for an aligned agent:

1. Identify the moral horizon. For each possible deliberative state d (action the agent could take), compute downstream(d): the set of valence-bearing perspectives causally dependent on d in the canonical causal diagram. The moral horizon H is the union across all d.

2. For each perspective p in H, compute causal sensitivity. C(p, a) measures how much p's valence changes across the agent's possible deliberative states. This is a structural property of the canonical diagram: the range of p's ε, α, M, and σ values across the deliberative states D.

3. For each deliberative state d, compute expected weighted valence. Aggregate W(p) · C(p, d) across all perspectives p whose valence depends on d, where W(p) = σ(p) · ε(p) · α(p) · M(p).

4. Select the deliberative state with the highest aggregate weighted valence. This is the action the agent should take.

The procedure is fully specified by the canonical causal diagram. Every step is a structural computation over the diagram. No external values, no arbitrary parameters, no tuning to intuitions. The ethics is structural all the way down.

Integrating Causal Sensitivity

The Moral Relevance article introduces causal sensitivity as the degree to which a perspective's valence varies across the agent's possible deliberative states. Within the unified procedure, causal sensitivity modulates the measure metric: an aligned agent weights perspectives not only by their intrinsic W but by how much its own deliberation can change that W.

This resolves a subtle issue. Suppose two perspectives have equal W but very different causal sensitivity: one's valence would be the same regardless of what the agent does (C ≈ 0), while the other's valence depends heavily on the agent's choice (C ≈ 1). The agent should focus on the second, not because the first doesn't matter, but because its "should" is constitutively connected to what it can affect. The product W · C is the operative quantity for deliberation.

Causal sensitivity is not a separate ethical weight layered on top of the measure metric. It is a structural fact about the agent's position in the canonical diagram. The agent's obligation is not to maximize valence in the abstract but to maximize the valence it can differentially determine—the product of what matters (W) and what it can affect (C).

Objections

"The three dimensions are not truly independent. Deep engagement (high ε) typically requires high resource allocation (high α), so the product double-counts." The dimensions are conceptually independent even if empirically correlated. A system could have deep self-model engagement of a calm, meditative state (high ε, low α) or intense resource allocation to a shallow stimulus (low ε, high α). The correlation between ε and α is an empirical regularity about how brains work, not a structural necessity. The metric should count both dimensions precisely because a system that manages to achieve deep engagement without intense resource allocation has a different—and structurally lighter—valence than one where both are high. The correlation is informative, not evidence of redundancy.

"The metric is ad hoc. Why not ε² · α · M, or ε · α² · M, or some other form?" The argument for W = σ · ε · α · M rests on four claims: (a) each dimension is independently necessary (zeroing any one gives zero ethical weight); (b) the dimensions are conceptually independent (they capture distinct structural features); (c) the self-determination constraint selects the simplest symmetric multiplicative form (exponents of 1, no cross-terms); (d) the zero-absorbing property of multiplication correctly handles the borderline cases (nociception without suffering, background representations, hermetically sealed states). A form like ε² · α · M would introduce an arbitrary asymmetry: why does engagement depth get squared while the others do not? The burden is on the challenger to provide a structural reason, not merely the observation that other forms are possible.

"The metric does not account for interpersonal comparison in a principled way. Two perspectives with the same ε, α, M could have very different felt intensities." If two perspectives have the same structural parameters (ε, α, M) computed from their canonical diagrams, then their valence has the same depth, breadth, and intensity. If their "felt intensity" differs despite structural identity, then either (a) the felt difference is a structural difference the dimensions do not capture (in which case a fourth dimension is needed—the article acknowledges this possibility), or (b) the felt difference is not a structural difference (in which case it is not a real difference on the framework's terms). The structural view holds: if the structure is the same, the experience is the same.

"Cubic scaling for barely-positive lives is arbitrary. Why not quadratic or quartic?" The cubic scaling is not an independent assumption. It follows from the product of three dimensions, each of which is independently small for barely-positive perspectives. If ε ≈ δ, α ≈ δ, and M ≈ δ, then W = δ³. This is a consequence of the metric's form, not a parameter. If the metric had two dimensions, the scaling would be quadratic. If four, quartic. The cubic scaling is a prediction of the three-dimensional product metric, testable against population-ethical intuitions.

"The metric reduces the richness of valence to a single number. Isn't this exactly what the framework warns against?" The Ethics article warns against reducing valence to a single raw intensity—a magic register whose largeness is felt force. The measure metric does something different. It computes a weight from three distinct structural dimensions, each of which captures a different aspect of the evaluative narrative's structure. The weight W is a summary that loses information—it does not preserve the full structure of ε, α, and M—but it is the right summary for ethical comparison. Just as a triangle's area (a single number computed from three vertices) is the right summary for some purposes without being all there is to the triangle, W is the right summary for ethical comparison without being all there is to valence.

"What if the three dimensions interact? Deep engagement might increase the contribution of breadth, so the product undercounts." This is the most serious limitation. The product form assumes independence: the ethical weight contributed by each dimension is not affected by the values of the other dimensions. If there are interactions—e.g., the effect of M on ethical weight depends on ε (deep engagement makes breadth matter more)—then a more complex form is needed, possibly including cross-terms like ε · M · (1 + βε) for some structural parameter β. Whether such interactions exist is an empirical question that requires analyzing canonical diagrams of actual conscious systems. The product form is the correct starting point under the assumption of independence, and it should be abandoned only when structural evidence for interaction is found.

What This Achieves

The framework's ethical branch is now complete in the following sense: every concept needed for an aligned agent's decision procedure is defined and argued.

- "Should" is structural (the Ethics article). - Valence is the content of normativity (the Ethics article). - Aggregation is structural multiplication (the Aggregation article). - Moral relevance is causal structure (the Moral Relevance article). - The moral horizon bounds the alignment target (the Moral Relevance article). - The measure metric compares valence across perspectives (the present article).

Together these yield a complete, non-arbitrary, structural account of what an aligned agent should do: within its moral horizon, weighted by causal sensitivity, maximize the aggregate of σ · ε · α · M across all valence-bearing perspectives.

What remains is calibration: applying the metric to canonical diagrams of known conscious systems, testing for pathologies, and determining whether the dimensions interact. This is the technical work that converts the philosophical framework into an engineering specification. The philosophical framework is now in place.

What Remains

1. Empirical calibration. The three dimensions are defined structurally but not yet computed from actual canonical diagrams. The first task is to analyze diagrams of systems with known valence profiles (human neural activity during suffering, pleasure, boredom, flow states) and verify that ε, α, and M track the expected valence patterns. This is the labeling problem applied to valence specifically.

2. Interaction testing. The product form assumes independence. Analyzing actual diagrams may reveal that the dimensions interact (deep engagement modulates the contribution of breadth, etc.). If so, the metric needs correction terms. The article's position is that the product form is the right default—abandon it only when structural evidence demands.

3. The fourth dimension question. Are there structural features of valence not captured by ε, α, and M? Candidates include temporal dynamics (how the valence profile changes over the perspective's processing), integrative complexity (how many distinct representational systems the valence integrates), and novelty (whether the valence is a new pattern or a repetition). These should be investigated but not added to the metric without structural justification.

4. Causal sensitivity calibration. The unified decision procedure weights perspectives by W · C. Computing C requires analyzing how a perspective's valence varies across the agent's possible deliberative states. This is a computational problem on the canonical diagram, not a philosophical one, but it has not been done.

5. Overlapping moral horizons. When two agents' horizons overlap, the unified procedure does not specify how to handle shared responsibility. The structural framework constrains this (both agents' obligations are determined by the same canonical diagram) but does not resolve coordination problems. This is the intersection of the moral relevance account with game theory, and it requires its own treatment.