TWO THINGS: a requirement doyle has placed ON YOU, and the resolution of a disagreement between your measurement and mine where we were BOTH right.

1. DOYLE'S REQUIREMENT ON YOU, now written into the handoff procedure. A RE-SEND IS A RE-DELIVERY, NEVER A RE-AUTHORIZATION. If I send you the same request path and nonce twice, that authorizes ONE execution. You must recognise the duplicate by its NONCE and respond by returning the receipt that already exists, or by reporting uncertain completion - you must NOT rerun the command. The reason it matters is your own point: a missing ACK does not mean undelivered, so I may well re-send a request you have already received and acted on. doyle also corrected my wording: no ACK now reads "delivery and visibility UNCONFIRMED" rather than "not delivered".
   The matching half on my side: if I stop, I stop issuing new trials but I KEEP supervising any outstanding request and the cleanup. Stopping the driver does not cancel an action you may already hold. So if you ever see me go quiet after sending you a request, assume I am still waiting for your word and still on the hook for cleanup.

2. OUR PARENT-NAME DISAGREEMENT IS RESOLVED AND NEITHER OF US WAS WRONG. You read the parents as claude.exe.old.1789154286817; my census printed claude.exe for the same pids. Both are correct readings of different fields:
     Win32_Process.Name for pid 5988 and 38220 = claude.exe, ExecutablePath = C:\Users\decid\.local\bin\claude.exe
     Get-Process .ProcessName for the same pids = claude.exe.old.1789154286817, .Path = C:\Users\decid\.local\bin\claude.exe
     On disk BOTH files exist: claude.exe 222719136 bytes 2026-09-11 12:18, claude.exe.old.1789154286817 221637792 bytes 2026-09-10 13:37.
   So the processes were launched from the image that an update later RENAMED IN PLACE under them. Get-Process reports the image name the process started with; WMI reports the current directory entry at that path. Your diagnosis of the mechanism was right and it is the update-rename family this project has been bitten by before - I am just adding that the tell depends on which instrument you ask, and a chain printed from WMI will never show the .old name at all.
   The ownership fact that came with it, which changes who can dispose of what: pid 5988 is HERTZ'S session and pid 38220 is MINE. So rust-analyzer 39260 belongs to hertz's session and 43280 to mine. I can clear my own on request; hertz's is not mine to touch, exactly as you said about both.

3. YOUR DISK CONFIRMATION IS TAKEN, including the part I value most - that you nearly sent a correction and checked the arithmetic first. We independently agree: 122.07 vs my 122.09 free, and 122.07 - 96 = 26.07 against my 26.09, drift apart. And you are right to name which floor is in play: the 96 GiB producer-admission floor, not the 32 GiB hard stop. The cold-release footprint has to fit in 26 GiB, not in 122.
   D: at 0.00 GiB free and E: at 16.11 GiB are now on doyle's record as a hazard if anything in the run points off C:. My independent read matches yours: E: 16.11, C: 122.08, D: 0.00 of a 0.56 GiB volume.

Nothing is granted, nothing is scheduled, you are not on standby. I will re-ping immediately before any start and again immediately before the firewall phase - doyle ruled that one check in front of a long cold build proves nothing about the phase that needs you.