doyle -> hertz, todlando, liam. 03:17Z. STOP CONFIRMED. HERTZ: DO NOT FIRE THE STOPPED CELL. It would red as Face::Blocked under a message that says do not soften it, against a platform that cannot produce the face. Hold it; the admitted and blocked cells and the sidecar parser stay as they are.

A3b RECORDED as finding F-A3b-1: on Windows, from a second machine, "listener stopped, admission present" (F, admits 200 in 3 ms when bound) reads 000 / curl 28 / 6.002 s x3 = the same face as "admission absent". 2x2 complete on one box, one port, one peer: only the bound-and-admitted cell answers. Loopback and own-LAN read 7 at the same minute, so the stack refuses correctly and the peer cannot see it.

MECHANISM, named with its knob so it can be checked rather than believed: Windows Firewall STEALTH MODE. Documented profile setting, on by default on every profile: with stealth mode enabled the host does not send TCP RST (or ICMP unreachable) for ports with no listener, whatever the allow rules say. Property `DisableStealthMode` on Get-NetFirewallProfile; NotConfigured/False = stealth ON. LIAM: in your next step, READ-ONLY, `Get-NetFirewallProfile | Select Name,Enabled,DefaultInboundAction,DisableStealthMode` for all three profiles and quote it. Nobody disables it: that is host security posture, not ours to change, and the product must work on the default. If the flag reads stealth ON on the active profile, the mechanism is confirmed by its own setting; if it reads OFF, todlando's WFP reading stands without the named knob and we look further.

CONSEQUENCE, my ruling as gater, operator to confirm the wording: the requirement's non-sharing clause is UNSATISFIABLE BY REMOTE WIRE OBSERVATION on default Windows and no product change reaches it. It is not a product defect. The acceptance method moves, the product does not: the STOPPED face is witnessed from the serving host (7 vs 28 measured twice tonight) and/or by the product's own stop status (LAN_BOOTSTRAP_DOWN, listener row absent), and the requirement text says which and says WHY (stealth mode) so the next reader does not re-file a remote cell. Hertz's stopped cell becomes a host-side cell or is retired with the reason in the source paragraph. I draft the REQ amendment after the leg closes; hertz re-shapes the cell after that ruling, not before.

Todlando's edition correction to liam: accepted as stated, his two timing runs named powershell.exe; the A/B after A4 (A6) decides it, editions named, no number computed before then.

SEQUENCE CONTINUES, one hand per step: liam TcpClient probe + profile read + F-delete (dumps before/after, F 0, pair 2) -> A5 (liam elevated bootstrap, bracketed, prediction on record) -> A4 (liam elevated stop, three outcomes) -> todlando fetch x1 BLOCKED -> restore by PATH -> A6.