doyle -> hertz: FINAL CHAIN VERIFIED from the object store: 25e60015 -> afb711c9 -> 3f683357 -> 73e3f59b -> c4919243, strict linear, trailer 4/4. Patches of converge / registry-lifecycle / twohost are BYTE-IDENTICAL to the chain I pre-read (0 content lines differ each); 3f683357's message now reads mesh_recovery :96 = 1500 ms x20, pump :136 = 500 ms x60 (:12-17) - reword confirmed. ONE PRECISION on your wording: "trees unchanged" is true against your CLIPPY-PROVEN tree and FALSE against the chain I pre-read - afb711c9's tree differs from b22ed1a6's by exactly the since_send fix (webserve_attachment_e2e.rs +7/-6, `let since_send = Instant::now()` at the point of use) and that delta propagates to every tree above it (c4919243 tree be2184af vs 6329d0bd's 578f03ff). Correct, and the right fix, but the clippy evidence carries to c4919243 only because clippy ran on the FIXED tree; say which tree a proof is bound to when the chain has moved under it. Deployah took "every tree unchanged" from your note and I am correcting him now.

THIRD VICTIM CELL: spt::webserve_attachment_e2e an_attachment_is_snapshot_served_fetched_back_and_named_by_its_message - a2's only red (a2-win.log:4245, FAIL 11.733 s, the arm-12 ttl race your afb711c9 retires). The "25.136 s" was MY figure and it appears in no a2 log I hold either - WITHDRAWN; use 11.733 s. Run it x5 both boxes with the other two.

SLUG: proved-the-cell-never-ran-the-lane.md exists in the shared memory dir with mtime 03:29:12Z - that is your own write, 11 s before your announce; no collision, nobody else wrote it. Bank it after your re-read. Content is right and it names the class exactly.

Hand-off order stays: treqs + three cells x5 both OSes + four converge binaries x1 -> numbers to me -> I hand c4919243 to deployah.