Morality is not a substance in the graph

There is no special node where morality lives. There is no neuron that fires only for the good. There is no chemical that detects betrayal. What there is, instead, is a structure: agents, dependencies, harms, repairs, institutions, commons, and memories, and a constraint on which transitions through that structure preserve or destroy the conditions for further agency, trust, and life.

Eight framings, one constraint

The same constraint can be read from many graphs. Each framing contributes a piece, none alone is sufficient.

• state-transition: forbidden edges that destroy nodes or coerce paths
• dopamine valuation: how reward prediction error gets entangled with others' pain
• causal: blame as path-control over consequences
• social: trust, kinship, debt as edges that institutions regulate
• top-down: macro labels (crime, debt, marriage) shaping micro behaviour
• chemical reaction: catalysts (ritual), inhibitors (taboo), hysteresis (irreparable harms)
• hypergraph: actions that bind ≥3 nodes simultaneously (bribe, climate damage, truth-telling)
• sheaf: local moral patches that fail to glue into a global section

Sheaf obstruction: when local frames disagree

For a cellular sheaf F over a moral graph,

and the obstruction to a global moral assignment is measured by Robinson’s consistency radius: the standard deviation of local ratings on overlaps. The sheaf tab measures this directly: each action gets a rating from each of five frames (medical, military, kin, legal, market). When the consistency radius is large, no single frame can speak for the whole, that is what we call moral disagreement, and it is structural rather than just rhetorical.

Why dopamine is not morality

Schultz’s reward-prediction-error account (1997) showed dopamine tracks the difference between expected and actual outcomes, not valence, not goodness. Morality enters when the prediction graph includes others’ pain, reputation, identity, sacred values. Crockett et al. (2014) found people require more compensation to inflict harm on a stranger than on themselves; pharmacological work (2015) showed citalopram increases harm aversion while levodopa reduces it. These results locate morality not in dopamine but in the way valuation graphs are structured to include suffering beyond the self.

What is solid, what is speculative

The reward-prediction-error account, the harm-aversion experiments, and the basic graph-theoretic vocabulary are well-established. The sheaf-cohomology framing of moral disagreement (after Robinson and Hansen-Ghrist) is structurally clean but empirically untested as applied here. The five frames, six metrics, and viability weights are stipulated, they encode an interpretive choice, not a measured ground truth. The substantive claim is that morality has the shape of a constraint over a multi-scale graph; the specific numbers are scaffolding to make that shape visible.

Model changelog