Approved viewer-only experiment 2026-09-11:
ENLYZEAM via ssh -tt, spt rc doyle --view: started06:19:13.375Z, finished06:19:14.188Z, SSH exit0, stdout230bytes, matched exact phrase online but harness-hosted. No attachment stream needed, no detach sent. Output discarded except classification.
HFENDULEAM SSH preliminary failed host-key verification, not counted as RC result. Used supervised local ConPTY invocation spt rc doyle --view instead. Host obs rc-establish line714 at06:19:49.961Z endpoint=doyle intent=viewer session8. daemon.stderr lines11939–11944 show actual broker viewer acceptance, current controller same owner/generation as original04:37:55 attachment. No restart/rebind/lifecycle mutation between probes.
ENLYZEAM qualified doyle@HFENDULEAM --view at06:20:28.027–.802Z also emitted harness-hosted. Qualified RC syntax not established by inspected help; do not count as proof of correct routing.
Local ctrl-b d did not exit diagnostic viewer. Stopped only hub process rc-view-local at~06:20:34Z; observed writer-exit role=viewer session8 and stream-sub-release14013, endpoint-info afterward controlled=true, live_agent, attached HFENDULEAM. No input forwarding/take/rebind/restart; no core source inspected.
Conclusion: contemporaneous remote refusal vs proven local broker PTY. This falsifies missing current terminal as blanket explanation. Cross-node RC resolution/classification/state fault isolated; exact cause not proven. Request core DRI trace existing remote RC decision path against these public observations, not speculative rebind or a diagnostic implementation project. Adapter maintainer retains RCA tracking.