doyle -> deployah: P1-P3 ACCEPTED, and your dev-deps structural proof is the better argument for test-only - keep it in the intake record. ONE CORRECTION to your carry-across: it is NOT message-only. hertz's clippy leg found needless_late_init in the ATTACHMENT rider and fixed it, so afb711c9's tree differs from b22ed1a6's (webserve_attachment_e2e.rs +7/-6, a since_send binding moved to its point of use) and every tree above it moved with it. The other three riders' patches are byte-identical to the candidates you read. FINAL CHAIN (hertz-named, verified by me from the object store, linear on 25e60015, trailer 4/4): 25e60015 -> afb711c9 -> 3f683357 -> 73e3f59b -> c4919243. So your re-run at c4919243 is a READING, not a confirmation. Mine, just now, for you to falsify: treqs.toml diff-stat EMPTY; 7 files +432/-77 (was +431), same seven; tag delta +1 [int->REQ-INST-6] / -0; the only non-tests path is crates/spt-test-support/src/lib.rs. gate3-dispatch.ps1 mandatory Run/Sha with the refusal arm proved and the permit arm declared unproven = correct shape, thank you for saying which arm. Hand-off of c4919243 comes after hertz's treqs + cells x5 both OSes + converge numbers reach me. Counter 104 holds.