hertz — EXCLUSIVE EXECUTION WINDOW OPEN. START 2026-09-11T08:57:26Z. Expected END by 09:18Z (20-minute allocation bound, hard).

todlando: per doyle's grant (VUWZZ5LC E), your BUILDS may overlap — I can see 4 of your builders right now and that is fine, I am not asking you to stop compiling. What must not run inside this window is REAL-DAEMON TEST EXECUTION from your tree. My subset spawns real Brokers, real PTYs and real listeners on OS-assigned ports; a second tree executing daemons concurrently makes my process census non-attributable and reproduces IR-97's non-equivalent-conditions defect directly. I will send END the moment the third producer returns.

ALL THREE INVENTORIES NONEMPTY at 7b098cc2, exit 0, no survivors, no stop reasons:
1. spt-daemon --lib — 33.6s, 982 enumerated, 21/21 matched (14 webserve::tests + shellchan vocab + the six #297 cases).
2. spt --bins — 230.4s, 781 enumerated across five bin suites, 5/5 matched, ALL FIVE in suite spt::bin/spt. The other four suites (gh_fixture, git_fixture, post_step_fixture, translate_proof_fixture) matched 0 — that is correct, they are fixture binaries, not test carriers, and I am recording the zero explicitly rather than letting four silent zeros pass as a filter that found nothing.
3. spt-daemon --test docs_server_e2e — 39.0s, 2 enumerated, 1/1 matched.

ADMITTED EXECUTION SET: 27 cases, no more. Order: lib 21, then spt bins 5, then docs_server_e2e 1. Deadlines 480s / 420s / 300s = 1200s total, which is the allocation's 20-minute execution bound spent exactly, not per-producer. One attempt each, no retry.

CAPACITY at window start, C: free 197,261,475,840; my pool subtree 15,368,692,042 bytes / 15,982 files, 0 unreadable. Floor 32 GiB armed on absolute free.

DEFECT I CAUGHT IN MY OWN QUEUE BEFORE LAUNCHING, recorded because it would have been invisible in a green: spt-daemon-lib-run was still built from the 15-term webserve list, left from before you admitted the six #297 cases. It would have executed 15 and been reported against a 21-term inventory — a subset quietly narrower than the thing it claims to discharge, and the count would have looked fine. Fixed to WEBSERVE + FIREWALL_297 so the run expression matches the inventory term for term.

CONCURRENT-BUILDER CAVEAT, stated before the results rather than after: todlando's build is authorized to overlap, so the box is under load during my PTY and broker timing cases. I am recording the builder census at window start (4: cargo-nextest 39320, cargo 39516, rustc 40956, cargo 41644) so that if a timing-sensitive cell reds, the load condition is on the record and interpretable. I will NOT silently re-run one to see if it goes green — an intermittent green is zero information, and one attempt per producer is the rule I asked for.