Prove the Graft Golden File Worldline end to end

[flyingrobots]

Outcome

Prove that Graft can continuously retain and recover nearby workspace states through Echo and Keep without Git subprocesses in the storage hot path.

Dependencies

Blocked by #21 and #23. The import decision in #25 is informative but not required unless the accepted ADR says otherwise.

Scenario

1. Graft observes exact bytes for Foo.txt state A.
2. Echo admits the observation only after Keep stages, verifies, and retains the bytes.
3. An early insertion produces nearby state B.
4. Both states remain exactly recoverable by their logical identities.
5. Physical metrics demonstrate chunk reuse without making reuse part of logical identity.
6. Exact byte ranges load only overlapping chunks.
7. An external edit makes a stale-basis write refuse.
8. Process termination and restart recover a documented lawful state.
9. Corruption produces precise refusal.
10. Keep compaction preserves every logical identity and readable retained state.
11. Ordinary Git publication does not change Keep or Echo content identity.
12. Ignored workspace content follows Graft policy without leaking semantic paths into Keep identity.

Acceptance criteria

• The fixture runs against clean pinned Keep, Echo, and Graft revisions.
• No production storage operation spawns Git or Node per chunk, blob, or read.
• Receipts bind Echo identity to Keep verification and retention evidence without conflating their claims.
• Metrics report logical bytes, physical bytes, chunk reuse, write/read amplification, sync count, latency, peak memory, and restart time.
• Kill/restart and corruption matrices include every cross-repository durability boundary used by the scenario.
• POSIX and agent-native Graft reads return the same coordinate and exact bytes.
• The demonstration distinguishes implemented guarantees from watcher completeness or application policy limitations.
• The scenario is reusable as a cross-language conformance fixture.

Non-goals

• Replacing Git publication.
• Universal source-code semantics.
• Cross-organization Continuum transport.
• Cryptographic settlement proofs.