How Wajah names a likely cause, and how it checks whether the action worked.
Plain rules on public sensor readings. Every answer shows the clues behind it, and says when it does not know.
1 How a spike is detected
Each area's reading is compared with its own usual level (median over the previous 7 days) for that hour of the day. When PM2.5 stays well above that level (at least 90 µg/m³ and 1.5× baseline) for two consecutive hours, Wajah opens an incident. The incident ends when the reading comes back close to the usual level.
2 The five causes and their clues
Wajah scores each cause deterministically from live sensor readings and names the highest as “likely”.
3 How an action is judged
Three hours after a job is marked done, Wajah compares the sensor reading with the reading when the job ended. Then it subtracts the drop that spikes at that hotspot usually show on their own when nobody acts. What remains is the true net change.
4 The placebo test
To test how often this measurement method fools itself, we logged 24 fake work orders on real past spikes where nothing was actually done.
With nothing done at all, the method still claims success about 1 in 12 times due to turbulent weather noise. That is why Wajah always shows counts (such as “3 of 4 times”) rather than a single percentage in isolation.
What this cannot tell you
- ✕Clues are not proof. A likely cause is the best fit to current readings, not a finding of fact.
- ✕One result is weak evidence. Trust a pattern over multiple interventions, not a single reading.
- ✕Stations can be 1 km away. Sprinkling may help a street the station cannot see.
- ✕Work orders in this demo are simulated. They carry a DEMO tag. No real trucks were dispatched.