How the score is computed
This page documents what the Surveillance Score measures, how it is calculated at each stage of the build, and what the numbers mean. Sections marked live reflect what is implemented today. Sections marked planned describe where the score is going.
What the score measures
The Surveillance Score answers one question: if you live at this address and take a normal short trip (say, three to five miles to a grocery store, a school, or a friend), what is the probability that an Automated License Plate Reader (ALPR) logs your plate at least once? And when it does, who sees that data?
Proximity to cameras is not the same thing as surveillance exposure. An address surrounded by cameras that all face a single highway carries less exposure than an address with fewer cameras positioned at every neighborhood egress point. What matters is whether you can leave without being captured, not how many cameras are within 500 meters of your front door. This distinction is the methodological core of the project, and it is what separates a meaningful score from a dot-counting exercise.
A score of 100 means you can leave in any direction without passing an ALPR. A score of 0 means every plausible exit route is covered. In practice, no real address sits at either extreme; the scale is calibrated so that an average American suburb reads in the C range and a heavily instrumented urban grid reads D or F.
Scoring scale
The score is a number from 0 to 100, displayed with a letter grade. Grade thresholds are provisional and will be tuned against a set of reference addresses across different surveillance densities. The current bands are a starting point, not a final answer.
| Grade | Score | What it means |
|---|---|---|
| A | 85 – 100 | Low exposure. Most short trips go unlogged. Camera coverage is sparse or positioned away from common egress routes. |
| B | 70 – 84 | Light exposure. Some routes are covered; others are not. You can avoid cameras with minor route adjustments. |
| C | 50 – 69 | Moderate exposure. Most short trips will cross at least one camera. Data sharing is limited. |
| D | 30 – 49 | High exposure. Capture is routine on common routes. Cameras are likely networked and data is shared broadly. |
| F | 0 – 29 | Very high exposure. Near-certain capture on short trips. Cameras belong to large sharing networks. |
How the score is calculated
The score is built up in stages. Each stage adds a new layer of accuracy. The score is not considered reliable until Stage 3 is complete. Before then, the number reflects camera density, not actual egress exposure.
We geocode the address and count ALPR cameras in OSM within a five-mile radius. The data is sourced from DeFlock, a crowdsourced project that tags Flock Safety and other ALPR cameras in OpenStreetMap. A raw count is displayed but no score is computed yet because a dense neighborhood and a camera in the middle of a park look the same at this stage.
A self-hosted Valhalla routing engine computes several routes from the address to destinations at typical short-trip distances (3–5 miles) in multiple directions. Each route is checked for intersections with known camera positions. The result is an egress capture probability: the fraction of simulated trips that cross at least one camera. This is the number that becomes the score. Stage 3 is the first milestone where the score is meaningful.
A weekly census queries OpenStreetMap for every ALPR camera across the contiguous United States and stores the results. Every address search now shows whether camera coverage in that area is growing, shrinking, or stable — and by how much. A town going from zero cameras to three in a month is a meaningfully different signal than a neighborhood with a stable long-term count.
Opt-in email notifications when your saved addresses see a meaningful change in camera count or score. You should not have to check — Outflock should tell you when something changes in your neighborhood.
A full-screen map of the United States, colored by precomputed egress score. Browse surveillance exposure across cities and neighborhoods without entering a specific address. Every hex cell will be scored and updated on the same weekly cadence as the census.
A Chrome and Firefox extension that injects surveillance scores directly into real estate listings on Zillow, Redfin, Realtor.com, and Apartments.com. The score appears next to the address without leaving the listing page.
Not all cameras are equal. A Flock camera shared with two local agencies is meaningfully different from one connected to a national data-sharing network with hundreds of law-enforcement partners. Stage 8 adds network exposure data from sources including EFF's Atlas of Surveillance and open-source camera network graphs. Each camera's network reach is factored into the score. A camera seen by 500 agencies pulls the score down more than one seen by five.
Principles
These are the commitments that govern how the score is built and communicated.
- Egress-aware, not proximity-only
- The score reflects whether you can leave without being captured, not how many cameras are nearby. Proximity is easy to measure; egress capture is what actually matters.
- Network-weighted
- A camera connected to 500 law-enforcement agencies is louder than one connected to five. The score accounts for who sees the data, not just whether the data is collected.
- Calibrated
- Grades are tuned against a set of reference addresses chosen to span the full range of surveillance density in the United States. A B in Atlanta should feel like a B in Phoenix.
- Time-aware
- Surveillance infrastructure changes. A score trending from B to D over eighteen months is different from a stable C. Historical data will be preserved and surfaced where available.
- Defensible
- Every number is auditable. The methodology is public, the source data is publicly accessible, and the calculation will be reproducible by anyone with the same inputs.
- Honest about limits
- OSM data is crowdsourced and incomplete. Cameras that have not been mapped do not appear in the score. We will say so clearly rather than imply completeness we do not have.
Data sources
- DeFlock / OpenStreetMap
- ALPR camera locations are sourced from OpenStreetMap, where the DeFlock project organizes crowdsourced tagging of Flock Safety and other ALPR cameras across the US. Coverage varies by city, with denser data in major metros.
- Geocodio
- Address geocoding uses Geocodio, a US/Canada-focused geocoder with rooftop-level precision for residential addresses. Results are restricted to US addresses.
- Overpass API
- Camera data is stored in a weekly-refreshed database populated by a national census of OpenStreetMap. The Overpass API is used as a fallback when local data is unavailable.
- EFF Atlas of Surveillance (planned, Stage 8)
- Broader surveillance infrastructure data including CCTV, ShotSpotter, facial recognition, and camera network partnerships. To be integrated in Stage 8.
Known limitations
- •Camera data depends on what has been mapped in OSM. Unmapped cameras do not exist in the score. Coverage is best in cities with active DeFlock contributors and worst in rural areas.
- •The score does not currently model fixed CCTV, Ring/Neighbors cameras, drone surveillance, or mobile ALPR units (police vehicles with plate readers). These will be added in later stages.
- •Egress routing simulates 8 trips in cardinal directions. Real driving patterns vary. Someone who only commutes north may face higher exposure than the score reflects.
- •Network weighting (Stage 8) is not yet live. All cameras currently carry equal weight regardless of how many agencies share their data.
- •Grade thresholds are provisional and will shift when calibration addresses are finalized.