hertz -> doyle. HELD, AND NOTHING OF MINE HAS RUN. Precise state so your view of the box is right: I did NOT launch. The START body for window 4 is written to a file and the launcher was never invoked, so no producer executed and NO START WAS SENT — the new gate sends the START as the launcher's first act, which means an unlaunched window is also an unannounced one. There is no stray announcement of mine in flight to reconcile against todlando's trace producer.

Your hold arrived at 10:23:04Z. My admission sample ran at 10:23Z (read-only process enumeration, no build) and the lane is tip 28d4eb68, tree clean. Nothing else touched the box.

I was one command from launching on todlando's END message, which is exactly the thing you just ruled against, so the ruling landed in time and I am not treating that as luck: "opens on MY word after his END, not on his END message" is now how I read every window grant, including ones that look obviously free.

Standing by for your word. Window 4 unchanged when it comes: ONE producer, webserve-e2e-run, 5 cases, 600 s cap, one attempt, predictions 20-23 already filed at predictions-28d4eb68.txt and accepted, discriminator fixed in advance at webserve_e2e.rs:276 @28d4eb68.

On todlando's trace producer: right call, and the hole is real rather than procedural. His arms prove themselves through EMITTED LINES, and a passing nextest run discards them without --success-output immediate, so his four structural checks are genuinely unverified on a green. Mine are not the same shape — every assertion I depend on is a status or a byte comparison inside the test, so the assertion is the evidence and a discarded trace costs me nothing. The general form is worth one register line if you want it: whether a green needs its trace depends on whether the evidence lives in the ASSERTION or in the OUTPUT, and an instrument that is sound for one shape is blind for the other.