🟠 LOW SEV · Inference Unlocks Without Sessions 🔥 Phones are unlocking but no foreground sessions are starting <|📈 graph> · <|📊 service dashboard> · :rotating_light: alarm definition · :no_bell: silence _LOW SEV, and treat as PROVISIONAL. Catches 'consuming fine but predicting nothing' - no transport metric covers it. Screen time flatlines at zero while every other signal looks healthy, so a parent opens the app and sees 0 minutes for a kid who was on their phone all day. That is corrosive because the product looks like it is working. Rated LOW because this is a HEURISTIC, not a measurement: it assumes unlocks and session-starts stay correlated. That held across 3 days of prod (sessions run ~5-6x unlocks, both troughing to zero together overnight), but a legitimately idle-but-unlocked fleet would trip it. Watch for false positives before promoting. Unlocks are the denominator on purpose - a phone that unlocks should produce a session, at any hour. Overnight both go to zero together and this correctly stays silent: nothing is happening, so nothing is being lost. This replaced a 6h absolute-count window, which could not fire before the morning parent digest. NOTE: restarts and partition moves re-report every device's lock state once as a 'first observation', counting as a transition - so >= 5 is an arming floor, not an exact count._ Grafana vmain-bff6a80