Field note
A runtime lock that did not lock the runtime
A pre-live review found that version labels alone did not bind every result-affecting runtime surface.
How the claim changed
Belief
A frozen runtime lock should make a later qualification or run meaningfully reproducible.
Evidence
The first agentic-Jev lock froze several visible runtime identities and study inputs.
Contradiction
Pre-live review found additional result-affecting surfaces that were recorded incompletely or not rechecked before execution.
Correction
Before any new treatment outcome, the lock was strengthened to bind and recheck the host-side Jev implementation, TypeSafe SDK, Inspect/SWE runtime, Codex platform, container environment, resolved sandbox image, and benchmark harness. Live bridge qualification was tightened to exactly one Jev invocation and successful receipt.
Current claim boundary
Repository review is ready for the next live qualification gates. It is not evidence of completed v2 A/B/C adjudication or the full 24 × 3 pilot.