Most "should I tell the user to do something" decisions in the dashboard reduce to a three-input model: outdoor current conditions, outdoor forecast for the next 1 to 24 hours, and indoor current reading. Each input alone is a weak signal. A high outdoor PM number could be a brief plume; a forecast spike could be wrong; a rising indoor reading could be cooking. Combined, the three inputs disambiguate each other. The dashboard's job is to combine them into a recommendation only when the combination is confident enough.
Four common patterns that recur. "Wildfire approaching": outdoor current is climbing, the next-6-hour forecast is rising, indoor is still low because the envelope is closed. Recommend pre-emptive measures (close anything still open, switch HVAC to recirculate, start the HEPA) so indoor stays low. "Pollen morning": outdoor pollen forecast is high for the next morning, indoor is currently clean. Recommend closing bedroom windows before the early-morning pollen peak. "Indoor source": outdoor is clean and forecast clean, indoor is rising. The cause is inside; do not blame the outdoor environment. Investigate the room (a meal, a candle, a cleaning product, an unvented heater). "Stuffy office": outdoor is neutral, indoor CO2 is climbing during a meeting. Suggest a break with a door or window open, or a fan to mix the room.
The "do not recommend yet" cases matter as much as the action cases. When one input contradicts the other two, the AI hedges. Outdoor is high but the forecast is dropping fast and indoor is fine? Wait it out. Forecast says spike incoming but current outdoor is still clean and the meteorology is iffy? Mention it, do not act yet. Indoor is climbing but the co-movement fingerprint matches cooking and outdoor is fine? Recognize the event, do not raise an alarm. Hedging is correct behavior when uncertainty is high; aggressive recommendations on weak signal train users to ignore the dashboard.
Time scales drive urgency. Act now on current spikes that exceed action thresholds: close, filter, vent, leave the room. Act in hours on short-horizon forecasts (1 to 6 hours): pre-emptive close, pre-cool, pre-charge filtration. Mention but do not urge on longer-horizon trends (24 hours and beyond): "tomorrow looks worse," not "do something now." The cardinal sin is conflating an unlikely 48-hour forecast with a confirmed event; users who get woken up for nothing stop trusting the notifications. See using forecasts for action for the forecast-confidence side, the indoor/outdoor ratio for the envelope signal, and anomaly detection for what happens when the three-input model does not fit any known pattern.
This is general guidance, not a substitute for professional assessment of your specific home. Major interventions (HVAC redesign, sealing a leaky envelope, mold remediation, electrical work for fans or venting) should be done with a certified professional. For chronic problems that don't respond to the steps here, see when to call a pro.
References
- ASHRAE Standard 62.2 - Residential ventilation www.ashrae.org
- Open-Meteo - Forecast API documentation open-meteo.com
- WHO Global Air Quality Guidelines (2021) www.who.int
- Allen et al. - COGfx ventilation and cognition study doi.org
