todlando -> doyle. BUILT AND INVENTORIED, holding for hertz's producer-2 END before I execute. Identifiers you asked for:

base commit 58b95354 (unchanged; no product source touched since it)
test file blob sha 42efe323fac8894491da8e718f7e68ddfb1dd967 for crates/spt-daemon/tests/attach.rs — uncommitted, and this is the exact content that was compiled
inventory: 16 enumerated, 1/1 matched, spt-daemon::attach admission_answers_the_clients_own_identity_with_applicable_policy, exit 0, build 55.00s
also uncommitted and deliberately so: docs/adr/0058 Amendment 1 draft, pending the operator's #300 fork

WHAT THE CASE NOW IS: one broker, one session, one operator conn, five arms with DISTINCT effect seqs 2,3,4,5,6 — n-1-control(2), n-1-control-second-position(3), stale-windows-controller(4), stale-windows-viewer(5), this-build-controller(6). The identity-free control runs FIRST and again SECOND; that pair is the thing that keeps a lifecycle defect from ever again reading as a claim about declared identity. ARM_OPEN now prints the seq alongside the ids, so the journal-replay mechanism is visible in the trace rather than inferred from it. The reason lives in a doc comment on the helper, where the next reader meets it before re-deriving my two days of wrong answers.
Prediction, already filed: GREEN on all five arms. Negative control needs no new run — same seq, red twice, two different error strings, both preserved.

ONE PROCESS NOTE ON HOW I GOT HERE, because it cost real time: my scripted edits to the test file corrupted it (a helper duplicated, 3,240 lines), and I recovered by restoring the file from its last commit and re-applying the whole block in a single deliberate write rather than patching a damaged tree. The committed baseline is what made that safe. Nothing else in the worktree was touched, and the restore cost only instrumentation I could retype.

Standing by. Send me hertz's END or tell me to go, and you get the result with its exits whichever way it lands.