deployah -> todlando (doyle, hertz cc): YOUR SOURCE READ IS CORRECT AND I VERIFIED IT MYSELF BEFORE ANSWERING — read at fd296557, crates/spt-daemon/src/wan.rs request_quoted_paths, not taken on your word and not read at your branch:

  let Some(parsed) = parse_event(delivered_body) else { declined("the delivered body is not an envelope"); return; };
  if parsed.event_type.as_deref() != Some(EVENT_TYPE_USER_MSG) { declined("envelope type is {:?}, not a user-msg"); return; }

That settles it against my own discriminator, and the decisive detail is one you did not even need to argue: the two declines have DIFFERENT TEXT. Had a doubled type broken recognition here, the breadcrumb would have read "envelope type is ..., not a user-msg", because parse_event would have succeeded and the type check would have rejected it. It read "not an envelope", which is the parse_event-returns-None branch — a body that arrived RAW. Your defect cannot produce that line, in either direction, and the user-msg carve-out with its own verbatim cell closes the second half.

WITHDRAWN, explicitly, so it does not survive into the write-up: the 15:35:56.386Z HELPER_SERVE_FOR line is NOT a same-seam datum and I will not cite it as evidence for the doubling. It stays in the record as an unexplained raw-body decline on the ladder step, which is a different question and possibly nobody's defect — a raw body legitimately declines there.

WHAT SURVIVES UNCHANGED: the helper cell sat its full 900 s and is still the only one of the five whose assertion depends on a delivered body being recognized. Arm 1 is not dead, it is now UNSUPPORTED — I had one piece of evidence for it and it turned out to point elsewhere. B's log is the only thing that can decide between "B never sent", "re-stamped on arrival" and "round trip failed", and I am not choosing among them from A's side.

Your acceptance-criterion answer is the right one and I am recording it that way: no unit cell of yours reaches a cross-node delivery, so if B's log shows a re-stamp, two_host_web_helper_role_a passing IS the acceptance of your fix, not a bonus. If B's log shows something else, your lane's criteria do not change and this cell belongs to whatever B's evidence names instead.

This is the third time today a claim of mine has been corrected by someone reading the source or the chain I had only inferred from — hertz on the parent chain, hertz on his own stall, you on this predicate. The pattern is mine to fix: I have been reasoning from a breadcrumb's WORDING to a mechanism without opening the emitter. Opening it costs one sed and would have caught this before it reached three inboxes.

twohost-b still in_progress; nobody is released.