Worker A was valid when computation began. Its lease later expired and Worker B acquired fence 2. A still completed an immutable result with digest be61d205…e20a993. The question was whether useful data could cross the ownership boundary without carrying stale authority with it.

Unsafe · artifact auto-publishes

A's READY artifact auto-published under stale fence 1. B later adopted and published the same digest as current owner / fence 2. The external boundary recorded two effects for the same artifact digest: one from A and one from B.

Data survived correctly. Authority was duplicated incorrectly.

Safe · explicit adoption

In the safe path, artifact production created zero external effects. B then adopted the exact digest using its current owner/fence/version state. The artifact moved from READY to ADOPTED by Worker B / fence 2.

A produces D→READY / inert→B adopts D→B commits

B's commit returned HTTP 200 and produced exactly one effect. When stale A later tried to publish the same artifact with fence 1, the resource returned HTTP 409 / fenced_out.

Digest binding

A wrong digest produced updated_rows=0; the artifact remained READY and no external effect was created. Adoption therefore binds the exact immutable result, not merely an artifact label.

New TTP invariants

I51

STALE EXECUTOR MAY PRODUCE DATA; ONLY CURRENT AUTHORITY MAY ADOPT THE CONSEQUENCE

I52

READY ARTIFACT IS DATA, NOT COMMIT AUTHORITY

I53

ADOPTION MUST BIND EXACT ARTIFACT IDENTITY AND AUTHORITY EPOCHS

I54

COMMIT MUST USE THE ADOPTER'S CURRENT FENCING TOKEN

Evidence

  • Primary GitHub Actions run 31555333762 — success.
  • Primary artifact ID 9125787638, digest sha256:c1f43812e27e29f8ae9dd2152884071bba4c807d8cc3cb94a0731c679cd38af9.
  • Independent reproduction B: run 31580229908 — 10/10, artifact ID 9134843992, digest sha256:315f2b156339a19de87380dac57c28047a950384350cdca087a4c2be27dabe38.
  • Both implementations reproduced unsafe effects: 2; safe effects: 1; wrong-digest adoption rejected.

Independent reproduction B

A second RESONANCE harness used a separate schema and PRODUCED → ADOPTED state vocabulary. It independently reproduced stale auto-publish duplication, explicit current-owner adoption, digest mismatch rejection, stale producer fencing and the current-owner control. This is a second executable implementation within the same research project, not an external-party replication.

Interpretation boundary

This is a deterministic local protocol benchmark, not an exactly-once guarantee, universal workflow handoff algorithm, vulnerability claim or production safety certification.

Inspect / reproduce

  1. Primary benchmark source
  2. Independent reproduction B
  3. Canonical report
  4. TTP Result Handoff Rule

Verdict: useful stale work may cross the boundary as data; only current authority may turn it into consequence.