The interpretation layer treats weekday and weekend as separate priors. Most households spend Monday through Friday daytime at school or work, returning in the evening, and Saturday/Sunday at home. The air-quality pattern follows: weekday days are low-occupancy and low-event, evenings spike for cooking and cleanup, weekends are higher-occupancy throughout with concentrated meal and cleaning events.
CO2 tracks the schedule almost exactly. A weekday-empty home settles within a few degrees of outdoor CO2 by mid-afternoon; on weekends it spends much of the day at 800-1,200 ppm depending on activity. VOC patterns shift toward weekends: people cook more, clean more, and host more. PM2.5 events from cooking cluster on weekend evenings and Saturday lunches in many households.
The dashboard learns each household's schedule from the first few weeks of data, then uses it as a Bayesian prior when interpreting events. A 1,400 ppm CO2 reading on Tuesday at 2 PM is unusual (suggests someone is home unexpectedly, or the HVAC is starved); the same reading on Sunday at 2 PM is expected. The dashboard's "anomaly" flag respects the schedule: same parameter, same value, different anomaly classification.
Remote and hybrid workers break the standard pattern. A household with a full-time work-from-home occupant shows weekday-daytime CO2 profiles closer to weekend patterns; an empty-house pattern is the anomaly. The dashboard's learned schedule handles this without explicit configuration: the prior shifts with the data. Holiday weeks, school breaks, and travel weeks produce their own distinct patterns the dashboard recognizes after a few exposures.
References
- Klepeis et al. - National Human Activity Pattern Survey doi.org
- LBNL Indoor Air Quality science portal iaqscience.lbl.gov
- EPA - Improving indoor air quality www.epa.gov
- Persily - Indoor COâ‚‚ and ventilation doi.org
