{"body":"Docs-only: one INFRA-REGISTER entry (IR-92) and the day's GATE-W2-272-CHECKLIST rows.\n\nIR-92: MSG_IN has two publish sites (delivery.rs:757 poll_drain, startup.rs:1064 the api-listen relay sink). The spt-hosted inject leg (inject.rs:75 try_spt_hosted_inject, reached by the send-time direct inject, --force-native, the idle-edge/parked/native drains and WAN ingress) delivers with no publish anywhere. Hook-hosted live agents: found busy -> spool -> hook poll -> MSG_IN; found idle -> inject -> wake -> nothing. Measured 2026-09-09 on hfenduleam across both agents' full io-events streams: SENT-to-idle 11/11 unrecorded on the receiver, QUEUED 8/8 recorded. REQ-IO-MSG-EDGES names the inject edge in its own title. Remedy: one observation-only publish at the Some((true,_)) arm of try_spt_hosted_inject. Fix lane unassigned (product: todlando; test-side: hertz).\n\nRides the next docs batch to main; a push to main runs unit on both runners.\n\nCo-authored by: doyle\nCo-authored by: hertz\n\n🤖 Generated with [Claude Code](https://claude.com/claude-code)\n\nhttps://claude.ai/code/session_01PL7KPiGpCSvJMd5E65DKfq","mergeCommit":{"oid":"79ac8d045b3cf6889ce94c43aaff1a46624fa930"},"mergedAt":"2026-09-11T01:20:35Z","state":"MERGED","title":"docs(register): IR-92 — the message that wakes an idle endpoint rides a leg that never publishes MSG_IN"}