tenant.openagents/omega

Publish the proof harness's Sarah records on the kind the host reads (omega#49)

Publish the proof harness's Sarah records on the kind the host reads (omega#49) The device journey on `ios-simulator (iPhone 17 Pro, iOS 26.5)` showed owner sends working end to end — `OWNER SEND ACCEPTED · message_seq 1 · turn.1`, both messages projected back and rendered as confirmed entries — and Sarah's reply never arriving, even though the host printed `SARAH REPLIED 6feacf9ca0b62ff8 · turn.2`. `live_device_proof_host` published Sarah's half as kind 14, a bare NIP-17 rumor. The host subscribes to gift wraps addressed to itself, the Sarah record kinds `[24200, 44200, 44300, 44301]`, discovery, and the NIP-AE/RS/ER kinds — a plain kind 14 is none of those, and `admit_event` would drop it through its final `else` even if it were delivered. A Sarah turn is kind 44300. So the harness was publishing on a kind its own host is built to ignore. Every reply it reported sending was discarded in transit, which means the reply arm of this proof had never once run, and the seeded greeting had never been projected either. Triangulated by the previous run: `projectedSourceEventIds` held only the two owner events, while the seeded engram on kind 30174 *was* read and quarantined — so the transport was fine and the kind was not. All three harness publishes move to 44300: the reply, the seeded greeting, and the deliberately unreadable quarantine source. That last one matters for the same reason — a source the host never receives cannot be quarantined, so on kind 14 it proved nothing about quarantine either. Nothing in the shipped host changed. This is the harness learning to speak the protocol the host already implemented. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>