SadakSense: road-quality evidence from the phone in your pocket
Why it exists
The road remembers every pothole. The ward does not. Road condition is measured, when it is measured at all, by occasional surveys that cost a great deal, cover little, and are never seen by the people paying for the roads. Meanwhile every phone on every dashboard is already feeling every bump.
SadakSense is built to be a public instrument of accountability — not navigation, and not a research toy. The citizen who contributes the data is the same citizen who gets to use it.
How it works
On the phone. During a trip the app samples the accelerometer a hundred times a second alongside GPS, rotates the signal to true vertical, and cuts it into one-second windows it can describe — how rough, how sharp, at what frequencies — gated by speed. It picks out potholes and speed-breakers by their shape. The first and last hundred metres or so of every trip are trimmed, so a trip does not point at anyone's door. Raw recordings leave the phone only if you opt in, and are deleted within seven days.
On the server. Each window is matched to a road segment, and each vehicle is calibrated against itself, so a stiff hatchback and a soft SUV can be compared. Passes are combined statistically, segment by segment, with an honest measure of how sure the estimate is. A pothole counts only when other devices confirm it.
What comes out. A confidence gate decides what the public sees: if the evidence cannot stand behind a segment, citizens never see it. Above the gate come an Infrastructure Health Index for an area, monthly report cards, and change detection that notices new damage and new repairs. And for any segment that clears the gate, the app shows which body is responsible and drafts an evidence-backed grievance. Below the gate, it refuses to.
Standards, stated carefully
Every output speaks two languages: a colour a councillor can read and a metric an engineer can check. The engineering metric is the International Roughness Index. A physics-based engine that follows the ASTM E1926 reference quarter-car is running in shadow alongside the current model. What is not done yet is calibration against a surveyed profile of a real road — so today the numbers are defensible for ranking roads and tracking change over time, and not yet as absolute IRI. That is the next milestone, and it needs fieldwork, not more code.
Where it stands — September 2026
- Live on Google Play and the App Store (1.0.22, 30 August), first approved in July.
- A government portal at portal.sadaksense.com, alongside the public site and an operations console.
- Road networks loaded for Delhi-NCR, Bangalore and Mumbai, with assembly constituency boundaries for all three states.
- Next: the calibration fieldwork, ward boundaries, and the names and grievance channels of the representatives each segment answers to.
What has been hard
"Implemented" and "running" turned out to be different words. For months production held no data at all, because the app never registered the device it was running on. Once data flowed, one phone re-sent the same chunk seventy-four times — and since confidence grows with the number of passes, the most duplicated roads became the most "certain" on the public map. Worst of all, the filter meant to discard readings taken while someone was handling the phone was throwing away more than half of every drive, and mostly the rough half, because a real shock from the road looks a lot like a hand picking up a phone. Each of those would have produced confident, wrong numbers. Finding them is the job.
Comments
No replies yet. When someone answers you, it lands here.