Where the person-state and voice programs stand

The test program

M0 Arena is active, building a sealed longitudinal comparison against raw history, summaries, cheap counts, learned memory, and larger models. M2 text perception has started early. The later architecture and decision milestones remain locked behind their gates.

Sim-C detected a planted exact-state effect in a synthetic arena on October 10. That is a useful check of the arena, not a real-person decision result or a scale win. No participant contact or real diary enrollment has been cleared.

A separate speech path

Affective Voice now has an offline perception pipeline that can produce a versioned SpeechObservation with transcript, acoustics, quality, uncertainty, and provenance. An initial TTS adapter exists. Core integration, streaming, validated production speech-emotion inference, and human listening evaluation are not done.

Voice observations are evidence about an utterance. They do not become a person's state by themselves, and a speech pipeline result does not establish that audio improves a Core decision.

What we can offer first

The proposed first commercial discussion is evaluation of existing agent transcripts for missed escalation, lost context after topic shifts, and inappropriate agreement. A production evaluation service or live state API has not shipped. We will keep the distinction between proposed product, implemented tooling, and validated outcome visible on the site.