Vima Brosta · Surveillance & Privacy Lab · Program 01
Offlock is a navigation instrument that plans your route around the license-plate readers mapped across the US and Canada. It computes on your phone, shows you every request it makes, and never asks who you are. Launching on iOS in late 2026.
The lab
The Vima Brosta Surveillance & Privacy Lab studies how automated surveillance is built, bought and searched, and builds instruments that hand a measurable share of the street back to the person on it. Offlock is the lab's applied program: three lines of work, one product name.
Surveillance-aware routing on OpenStreetMap. Every mapped reader, which way it faces, and a route that prices it honestly against your time. Computed on the phone inside downloaded regions.
S01: a palette-locked camouflage computed against the person detectors behind video search. Two colourways, Night and Fog, scored on detectors they were never tuned on.
Off-Grid: whole regions on the phone. The map, the routes and the readers, downloaded before you need them, for the day the signal goes dark.
Position
Maps of plate readers already exist, and they matter. DeFlock and thousands of OpenStreetMap contributors put the readers on the map; the lab's contribution starts where the map ends.
A map tells you where the cameras are. An instrument changes what you do about them. Offlock treats each reader as a cost with a direction, prices it against your time in the open, and lets you pick the trade. It never claims a route is clean, because the map is never complete. It shows the counts it can stand behind.
The same discipline runs through everything the lab publishes. Every film cites its sources on screen. Every S01 number says which detectors it was scored on and at what distance. Every network request the app makes is listed inside the app.
Where OpenStreetMap records an operator, the largest fleets of mapped readers are retail, not police.
Operators as recorded in OpenStreetMap, September 2026. Most readers have no operator recorded.
Interactive · Real roads, real readers
Any two addresses in the US or Canada. The fastest route is what an ordinary GPS gives you. Offlock prices every mapped reader on the way and offers the trade, first Balanced, then Strict. Same router, same map data as the app.
Counts include only readers mapped in OpenStreetMap, within 40 m of the road and, when the facing is mapped, pointed along it. Coverage is only as complete as the map. No live traffic.
Live map
Loads street tiles from OpenFreeMap when you start. That's the only request this site ever makes to anyone else, and it carries no address: just the tiles you look at.
Method
The engine is Valhalla, the open router that runs on OpenStreetMap. Offlock adds a planner on top that treats surveillance as a generalized cost. Five steps, the same on the server and on the phone.
Valhalla computes the fastest route and up to two alternates, with your toll, highway, ferry and unpaved preferences.
The on-device dataset is scanned along each corridor. A reader counts if it is within 40 m of the route and, when its facing is known, pointed along that road.
The router is asked for new routes with those specific readers excluded. For directional readers only the watched road is closed. This repeats a few times.
Every candidate is scored by travel time plus λ per reader passed, under a detour cap. The fastest route always stays on the list.
Each option shows its computed deltas. Labels like “Fewest readers” appear only when the data supports them, and no route is ever called clean.
Camera data reaches routing through one audited adapter, and only when avoidance is on. The map overlay is informational and cannot influence routes. Driving behaviour never influences routes either. Architecture tests enforce these boundaries.
Program 01 · The instrument
Search, alternates, turn-by-turn guidance, rerouting, speed limits, CarPlay in testing. And on every screen, the number the others don't print: how many readers this trip passes.
Three routes, each with its time, distance and reader count. “Routed on this iPhone. No routing request left it.”
Maneuvers, speed limit, reader and enforcement alerts ahead, spoken guidance, background guidance and dead reckoning through tunnels.
Readers passed against the fastest route offered, and a weekly total. Counts only, kept on the device for six months, off with one switch.
| Feature | What leaves the device | Where it goes |
|---|---|---|
| Routing | Start, destination, preferences and the readers to exclude. Nothing when the region is downloaded: the route is computed on the phone. | Valhalla via the Offlock gateway, which never logs bodies or coordinates |
| Search | The query text and your position rounded to about 1 km for ranking. | Geocoder |
| Camera data | Nothing about you. The whole national dataset is downloaded and every lookup happens on the device. | Static file host |
| Driving Score | Nothing. Telemetry, events, trips and scores are computed and kept on the device. | Nowhere |
| Analytics | Off by default. If you opt in: event names and coarse buckets, never a location. | A separate endpoint, separate connection |
| Idle | Nothing is sent while the app is in the background, and GPS turns off unless guidance is active. | — |
#mALPRactice · Findings
The lab publishes its findings as vertical films built from reporting, court records, procurement documents and vendors' own materials. Each card cites its source. The on-screen text in every film is in the app's own words: no route is ever called clean.
Films are 9:16 and load only when you press play. Camera data in every film © OpenStreetMap contributors, mapped by the DeFlock community.
Program 02 · Textiles · S01
Police searched Flock's video tools for people by clothing, not by plate. Other vendors sell the same search. S01 is a two-scale Voronoi camouflage, optimized on real pedestrian photos against three person detectors, with a fixed palette so it prints and still looks like clothing.
Fourteen macro regions of about 10 cm, for wide-angle cameras at 10 to 20 m, over 180 cells of about 2.7 cm. Each cell takes one ink from a locked palette. The tile is rendered onto the torsos of real pedestrians with the photo's shading kept, under random scale, rotation, foreshortening, fabric warp, lighting, occlusion, blur and distance, and the detectors are asked to find the person.
A third of the renders use a simulated 850 nm night view, which is why the palette carries two near-identical darks: a carbon black that stays dark to infrared and a graphite that goes bright. Colour caps keep any one “upper clothing colour” under 42% of the tile, because attribute search bins by colour.
It won't make you invisible. Nothing will. S01 is designed to make you harder to pick out by software, on the detectors it was scored against, at the distances we tested. It is not a face covering, and the hoods stop at the chin. Distinctive prints can make attribute search easier, and we say so.
Program 03 · Resilience · Off-Grid
GPS comes from satellites, not cell towers. Offlock already plans routes on the phone for downloaded regions. Off-Grid packs put the whole region there: the map, the routes and the readers, downloaded before you need them.
Works today: routing on the phone inside downloaded regions, with the same reader avoidance and nothing sent to a server. In development: offline map display, region downloads and an on-device search index, packaged as the Off-Grid tier.
Roadmap
Dates are targets, not promises. Subscribers hear first when one moves, in either direction.
The routing engine, guidance and receipts run daily on real streets. The camera dataset of every mapped US and Canada reader is rebuilt nightly from OpenStreetMap and DeFlock.
A small cohort of drivers in a few US cities. Subscribers are invited in the order they signed up.
Free. No account. US and Canada reader coverage from day one, with camera-aware routing, guidance and receipts.
Search, trip preview with reader counts, guidance, camera alerts and the arrival receipt already run in the Simulator. Waiting on Apple's navigation entitlement and head-unit testing.
Physical camera tests on fabric and people come first: 850 nm swatch tests, walk-bys at 5 to 20 m, on detectors the pattern was never tuned on. Then a limited run of H01, H02, T01 and T02.
Offline map display, region downloads with routing tiles and readers, and an on-device search index. Routing on the phone already works; the pack makes it whole.
Surveillance-aware walking routes are in design. Android is not yet scheduled; tell us below if that's what you're waiting for, and it will count.
Updates
One address. Progress notes when something real happens, a beta invitation when your turn comes, and nothing else. Tell us which programs you're waiting on; that's how we decide what to build next.
No tracking pixel in the emails, no analytics on this site, one-click unsubscribe. The address is stored once, in one place, for this purpose only.
