How confident the AI is when it makes a recommendation

Every recommendation the dashboard surfaces carries an implicit confidence level, and the wording is calibrated to that level on purpose. High confidence looks like a direct imperative: "outdoor PM_2.5 is 180 µg/m³, your windows are closed, run the HEPA on high." The data is unambiguous, multiple signals corroborate, and the suggested action is reversible if you change your mind. You should expect the AI to sound decisive only when those three conditions all hold.

Medium confidence looks like a prompt to investigate before acting: "the VOC index has been climbing for the last hour with no obvious driver, check whether anything was cleaned, painted, or unboxed recently." The signal is real (it has cleared the noise floor and persisted), but the source is ambiguous, and the right intervention depends on which source it is. The AI surfaces the candidates rather than guessing among them. See how anomaly detection works for the underlying mechanic.

Low confidence looks like an explicit hedge or a refusal to give specific advice. A single-channel spike with no corroborating signal gets "this might be a sensor glitch, wait 15 minutes and re-check"; see cross-sensitivities for why the SEN66 sometimes does this. A reading inside an unusual personal-context envelope (medical condition, pregnancy, immunocompromise) gets "this level is above the WHO guideline; your specific situation may warrant a stricter target, but I cannot determine that for you, talk to your doctor." The hedge is not the AI being timid; it is the AI being honest about what it does not know.

Confidence rises when multiple sensors agree, when the outdoor feed corroborates an indoor reading, when the pattern matches a known fingerprint (cooking, cleaning, wildfire infiltration, shower humidity), and when a published threshold is clearly exceeded. Confidence falls when only one channel moves, when the source could be three different things, when the user context is unknown, and when the reading is near a threshold rather than across it. The AI will never claim confidence on a medical interpretation, an individual prognosis, or a life-safety conclusion; those go to a clinician or to the dedicated alarm on the ceiling.

This is environmental information, not medical advice. The dashboard's readings help you make decisions about the air in your space. They do not diagnose conditions, interpret symptoms, or replace conversations with your physician. If symptoms persist, worsen, or coincide with a known exposure, talk to a healthcare professional. See the AI's medical-advice scope.