Trust Map

The Core Problem

99.2% of "cloudy" readings are bright

Out of 599 daytime "cloudy" labels, 594 had brightness > 100.
In Shenzhen, "cloudy" almost always means thin clouds that let light through.
Weather apps see clouds from satellite — they cannot tell thin from thick.

Brightness gap: rain vs dry = -1.2

Average brightness during rain: 110.2. Average when dry: 109.0.
The gap is -1.2 points — statistically indistinguishable.
In Shenzhen, you cannot tell if it is raining by looking at brightness alone.

Trust Matrix

What you want to know	Weather App	Your Window
Is it cloudy?	9/10 Satellite sees all clouds	2/10 Cannot detect thin vs thick
How thick are the clouds?	1/10 Cannot see thickness	8/10 Brightness = cloud thickness
Is it raining right now?	5/10 Knows system, not surface	9/10 RMS audio detects rain directly
Is it dark / gloomy?	2/10 Cloudy does not mean dark here	9/10 Brightness = actual light
Will it rain soon?	8/10 Weather system tracking	1/10 No predictive power

Weather App Your Window

Conflict Density by Hour

When window is bright (>100) but weather label says cloudy — the thin cloud blind spot

Hourly Rain Frequency

Percentage of readings where RMS > 30 by hour of day

Key Insight: Complementary Sensors

Neither source is sufficient alone — they are complementary

The weather app is best at predicting (will it rain soon?) and seeing cloud systems.
The window is best at detecting (is it raining now?) and measuring actual conditions.

The trust map shows they have inverse strengths:
App: good at macro/prediction, bad at micro/current
Window: good at micro/current, bad at macro/prediction

The optimal strategy is fusion, not replacement.

The Core Problem

Trust Matrix

Conflict Density by Hour

Hourly Rain Frequency

Key Insight: Complementary Sensors