An ownership_fact is persisted only as a PRESENT decision

Index: Known gaps

Status (2026-10-29): FIXED (schema v67). The round trip is pinned, and every FactStatus member now survives a dump and load.

What was lost, and the fix. A declaration's ownership_fact (ADR-075) was written only as a PRESENT triple inside the extraction_scope block. Schema v67 adds two keys: entity_ownership.statuses carries each non-PRESENT fact as [position, status, producer, diagnostics] beside the interned decision table, and ownership_without_scope carries the decisions of a snapshot with no recorded scope, without writing a scope (ADR-075 reads an unrecorded scope as unrecorded). A PRESENT decision encodes exactly as before.

Why no verdict changes. The only reader is model.extraction_scope.ownership_of, which answers None for every non-PRESENT fact and for a missing one. A decoded None and a decoded NOT_APPLICABLE therefore mean the same thing to every consumer in the tree. The loss is in the Fact object a direct reader could inspect, not in any decision.

Why the producer never hits it. extract/ownership_stamp.stamp_ownership is the only writer of ownership_fact. It writes Fact.present only, and it always sets extraction_scope in the same call. The no-scope and non-PRESENT states are reachable only from hand-built objects.

How it is pinned. tests/unit/storage/test_snapshot_field_roundtrip.py enumerates each ownership_fact field inside a scope, runs one case per FactStatus member, and names each lossy case (ownership_fact[<status>], ownership_fact[no_extraction_scope]) in KNOWN_LOSSY with this reason. test_every_known_lossy_entry_is_lossy fails when one of those cases stops being lossy, so the entry cannot outlive the defect.

Related. Advanced DWARF extraction status is the same pattern for dwarf_advanced.extraction, where the loss does change a predicate's inputs.