hertz -> doyle. RED in producer 1, classified, plus a rig defect of mine that halted the sequence. Window still open; I am NOT continuing without your word.

RESULT, producer 1 (spt-daemon --lib, 21 admitted cases), exit 100, 1.72s wall, 0.194s nextest, no survivors, pool_growth 0:
  21 tests run: 20 PASSED, 1 FAILED, 961 skipped.
  ALL SIX #297 bootstrap_firewall cases PASSED (0.020s-0.063s each). shellchan vocab PASSED. 13 of 14 webserve::tests PASSED.
  FAIL: webserve::tests::node_routing_reserves_facets_without_registry_fallback, panic at crates\spt-daemon\src\webserve.rs:1011, assertion left == right failed, left: 302, right: 200.

CLASSIFICATION — THE TEST IS STALE, THE PRODUCT IS RIGHT. This is my defect, not a product regression:
- Line 1011 is `assert_eq!(get(home.path(), "/LoCaL/").status(), StatusCode::OK)` — an incidental node-root assertion buried inside a FACET-ROUTING test.
- REQ-WEB-URL-NODE-PREFIX states the design: "A node root /<node>/ redirects with HTTP 302 to /<node>/docs/, preserving query parameters, for local and remote nodes." The product implements exactly that at webserve.rs:778-782 — remainder empty -> redirect to /<lowercased local_node>/docs/ with the query carried.
- I checked the arm is not reachable for a remote node before calling it correct: the `first.eq_ignore_ascii_case(local_node)` gate at 752 sends non-local labels to the peer/compat arms above, so `local_node` in that redirect IS the requested node and there is no local-docs-for-a-remote-root bug hiding behind my red.
- So #301 (a4017a15) changed the node-root contract, I updated the tests ABOUT the root — line 1218 already asserts FOUND for "/local/" — and missed this one assertion sitting inside a test that is nominally about facet reservation. Classic sibling-arm miss.

SWEEP DONE, not assumed: I grepped every node-root-shaped request across webserve.rs, webserve_e2e.rs and webserve_cross_node_e2e.rs. 1011 is the only survivor. And the selection is not hiding others — webserve::tests has exactly 14 cases and all 14 are in my admitted 21, so this module is fully executed, not sampled.

MY RIG DEFECT, reported because it cost the run and would have cost the next one: my driver printed each producer's stderr tail to a cp1252 stdout and died with UnicodeEncodeError on nextest's box-drawing characters. The crash happened AFTER producer 1's evidence was fully written — exit, observation.json, raw and stdout are all on disk and intact — but it killed the loop, so PRODUCERS 2 AND 3 NEVER STARTED. That is a labelled hole, not a pass and not a skip. Fixing the driver to write UTF-8 with replacement now.

WHAT I AM ASKING FOR, three things, no action taken on any of them yet:
1. Producers 2 and 3 (spt --bins 5 rc::tests; docs_server_e2e 1 case) are INDEPENDENT evidence axes — neither touches webserve routing. Your allocation stops later execution after a failure unless an independent axis is explicitly selected. I am selecting them explicitly and asking you to approve. They have never run, so this is a first attempt, not a retry.
2. The one-line fix to 1011: assert FOUND and location "/local/docs/". Test-side only, no product edit. Then a new frozen sha and ONE re-run of producer 1. I will not re-run it before the fix, and I will not call the current 20/21 a pass.
3. Confirm you want the window held open across this. It is 09:00Z, bound is 09:18Z. If you would rather I close the window, take the fix and re-run cold later, say CLOSE and I will send END now.

ON TODLANDO'S SIX CELLS INSIDE MY WINDOW: no conflict with this result. He ran them at 08:58:34Z; producer 1 ran 08:58:01Z-08:58:03Z and was already finished. His cells spawn no broker, PTY, daemon or listener by his account, and my producer's own census recorded zero survivors. Nothing about this red is attributable to his activity, and I am saying so rather than leaving the overlap as an unexamined confound.