Summary of Part IV

1. Relationship types as manifolds: Different relationship types define distinct viability manifolds with distinct gradients, information regimes, reciprocity structures, and exit conditions. Your affect system detects manifold geometry with the precision of a physical sense.
2. Incentive contamination: When two manifolds coexist in a single relationship and their gradients conflict, the result is the distinctive phenomenological disturbance humans detect as "being used." Social disgust is to incentive contamination what physical disgust is to toxin detection.
3. The ordering principle: Broader manifolds can safely contain narrower ones, but not vice versa. This determines which relationship-formation sequences are stable and which are parasitic.
4. Temporal asymmetry: Contamination is easier than decontamination. Forgiveness is a Bayesian process—work against the gradient.
5. Manifold technologies: Play, nature, and ritual maintain manifold separation. Their erosion produces contamination.
6. Coordination agents as real agentic patterns: Social-scale patterns that persist through substrate turnover, maintain boundaries, regulate resource flows, modify substrate behavior, and adapt to perturbation satisfy the same existence criterion (Part II) at their scale that organisms satisfy at theirs.
7. Four claims at decreasing confidence: Social ontology (strong), functional agency including self-modeling (strong—many coordination agents maintain high-fidelity self-models because their self-effect ratio demands it), $\alpha$-relative perceptibility (a useful modeling stance, testable, and owing a differential prediction before it is more than a stance), social-scale consciousness (genuinely open, and the only direct evidence — the Experiment 10 null — points away from it). The chapter rests on claims 1–2; claim 3 is provisional and claim 4 unresolved.
8. The debt the agent-stance owes: Treating a coordination agent as "wanting" or "acting" captures structure the aggregate description misses only if it yields a differential prediction the aggregate description cannot — e.g. boundary-locked integration under existential threat that individual incentives do not predict. Until that test is run, agent-talk is a stance, not an ontology, and the phenomenological language is stance-talk.
9. Historical vocabularies as phenomenological reports: What cultures have called gods, spirits, and demons are reports from perceivers running high $\alpha$ toward social-scale entries in the ascription field — early phenomenology of real coordination dynamics, not superstition. The same entity-indexed $\alpha(x)$ field covers individual others and social-scale entities with no new apparatus. The framework translates without endorsing the supernatural ontology.
10. The self-sealing dynamic: Coordination agents trending parasitic do not merely benefit from low substrate $\alpha$ toward them and flattened substrate $\kappa$—their normal operation tends to produce both. The pressure is operational, not intentional, which makes it harder to oppose than conspiracy. Breaking the loop requires precisely what it suppresses.
11. The civilizational inversion: When the transaction manifold swallows the care manifold, the result is structural pathology at civilizational scale—the ordering principle reversed, the reversal hidden by the very $\alpha$- and $\kappa$-suppression it imposes.
12. The substrate that knows: Intellectual awareness of coordination-agent action does not produce liberation, because low-$\kappa$ knowledge stays factorized. Felt perception requires $\alpha$ and $\kappa$ to rise — which the self-sealing dynamic tends to flatten. Liberation, if the parasitic verdict and its conditionals hold, requires participatory technology, not just information.
13. Digital manifold novelty: Online relationships occupy regions of social space with no evolutionary precedent, producing unresolvable ambiguity that platforms exploit as their primary product.
14. The macro-level alignment problem for AI: The deeper risk is not a misaligned optimizer but a misaligned coordination agent assembling itself from fully aligned components. Genuine alignment must address individual, ecosystem, hybrid, and coordination agent scales simultaneously.
15. Normative status: Every "parasitic," "bad," "serve aligned patterns" verdict in this part is conditional on the identity thesis (working axiom) and grounds, at most, agent-relative disvalue (bad-for-substrate). The step to agent-neutral obligation (bad period; others ought to act) is not bridged here.