Should locally inferred smart-home states be treated as data?

I’m broadly in favor of local-first smart homes: keeping motion, watch, phone, and microphone processing on the household network greatly reduces cloud exposure. But “local” does not mean “nothing sensitive was created.” A system that combines ordinary signals into “probably asleep,” “working,” “ill,” or “home alone” may produce a conclusion more revealing and actionable than any individual measurement.

I’d like to see an inference dashboard, not just sensor history. Each state could show its confidence estimate, expiration time, contributing sensors, model or rule version, and which automations consumed it. That would make it possible to spot a bad inference, revoke it, or delete it rather than leaving derived facts buried in logs. Confidence should be presented as uncertainty, not as a truth badge.

I’m not sure “ownership” is the right legal term, but privacy control should cover conclusions as well as inputs. Do inferred states deserve the same ownership and deletion rights as raw data?

The dashboard is the important part, especially the consumer list. A state can stay entirely inside the house and still be exposed to every automation, user account, backup, and debug log. I’d also want a way to disable inference by category, not merely delete individual records.

Confidence scores can create false precision. “0.83 probably asleep” looks scientific even when the model is just seeing a quiet house and an inactive phone. I’d prefer the evidence trail and a short expiry by default; confidence is useful only if the system explains what it actually means.

I’d separate rights from ownership. The case for access, correction, retention limits, and erasure seems strong when an inferred state is linked to a person, but calling it property may imply a legal rule that does not necessarily exist. The practical controls matter more than the label.

There’s a household consent problem too. One person may configure the system, while the inference concerns everyone nearby. A readable dashboard could make that social boundary visible instead of pretending the home has one user and one privacy preference.