Week ending October 4: from presence to audio events

The available journal for this week runs from September 28 through October 1. It starts with a room that would not become empty in Ash’s perception and ends with transcript events that remain distinguishable after a listener restart. No October 2–4 engineering work is asserted in this roundup.

The jersey-on-chair experiment gave a concrete explanation for false presence in the tested scene. State and behavior then became separate event-driven layers, and the three-service stack passed observed reboot and live-delivery checks. A recovery correction stopped restored history from masquerading as a fresh arrival.

Spatial development moved to one primary-target decision per frame. The qualified observer went live in advisory mode, while more ambitious continuity experiments stayed offline. Strong agreement between algorithms was not treated as proof that either had identified a real person correctly.

Audio made two different advances: Piper generated a verified voice file, and the native listener preserved a long spoken thought while adding runtime-based event IDs. Audible speaker output was not tested, and the listener was not integrated into cognition.

The week’s record also preserves deliberate stops. The September 30 microphone-array work stopped before flashing, and the October 1 log-write correction remained a local, unapplied proposal. Boundary coverage, recognition accuracy, failure handling, and broader recovery qualification remained work ahead.