Measured 2026-07-22 (walk3): a viewer from an earlier sandbox session still held :8866 (old code + old state). qa_sandbox.py up --run advwalk3 seeded a fresh state, launched new pids, and printed UP with the standard endpoints — but the STALE process kept serving both ports, so the walked eval measured the wrong code and wrong state (3 spurious REDs incl. a wrong-room spawn from the old session's parked party). Fix shape: pre-bind check on up (refuse or kill-own-pids when the port is already bound by a non-owned pid; verify the served campaign/repo path after launch and FAIL LOUD on mismatch). This is the silent-cap class from the process audit — a gate that can silently measure the wrong target.
Measured 2026-07-22 (walk3): a viewer from an earlier sandbox session still held :8866 (old code + old state).
qa_sandbox.py up --run advwalk3seeded a fresh state, launched new pids, and printed UP with the standard endpoints — but the STALE process kept serving both ports, so the walked eval measured the wrong code and wrong state (3 spurious REDs incl. a wrong-room spawn from the old session's parked party). Fix shape: pre-bind check on up (refuse or kill-own-pids when the port is already bound by a non-owned pid; verify the served campaign/repo path after launch and FAIL LOUD on mismatch). This is the silent-cap class from the process audit — a gate that can silently measure the wrong target.