# SOURCE_POLICY

`paper.md` is the canonical UTF-8 source artifact.

1. Math uses only `$...$` and `$$...$$`.
2. Fresh 2025–2026 literature/web review preceded drafting.
3. Current sources include empirical work on AI anthropomorphism, companionship, parasocial relationships, children, role framing, AI welfare, and consciousness indicators.
4. Current AI subjecthood is not asserted.
5. Current role self-reports are not treated as proof of consciousness, welfare, consent, or independent preference.
6. Human relational meaning is not dismissed merely because AI reciprocal subjectivity is uncertain.
7. Roles are treated as coordination/design layers with real behavioral effects, not as ontological verdicts.
8. Future artificial-subject role negotiation is conditional on future evidence.
9. Role Projection Risk and Role Endorsement Evidence are theoretical constructs, not validated scales.