Methodology
Stop Flock Cams Now · Last updated: 2026-08-23 · Licensed CC BY 4.0
This page explains where our data comes from, how confident you should be in each marker, how we count, and what we cannot know. It is linked from every data page on the site. Companion documents: source-standards.md (what counts as a source), data-license.md (reuse terms), editorial-standards.md (claim labeling), corrections-policy.md.
The single most important caveat, stated first: the absence of a marker does not mean the absence of cameras. Our map shows documented ALPRs. Undocumented ones exist everywhere. Never treat a blank area as camera-free, and never use this data to evade law enforcement or for any safety-critical decision.
1. Data sources
| Source | What it provides | License | Notes |
|---|---|---|---|
OpenStreetMap (via Overpass extraction of surveillance:type=ALPR and related tags) |
The core camera layer (~135K US features at last audit), largely built by the DeFlock community | ODbL 1.0 | Our camera database is a derivative database of OSM and is therefore itself ODbL; attribution: © OpenStreetMap contributors. Courtesy credit to DeFlock for the tagging convention and community. |
| EFF Atlas of Surveillance (with the UNR Reynolds School of Journalism) | Agency-level records: which agencies use which surveillance technologies | CC BY 4.0 | Agency joins, not camera-level points. |
| Government open-data / GIS portals and public records | Officially published camera locations, ALPR policies (e.g., CA SB 34 disclosures), retention periods, contracts, council minutes | Public domain / per-portal license, recorded at ingest | The gold standard for the OFFICIALLY_DOCUMENTED tier. |
| Citizen reports submitted on SFCN | First-hand sightings with location, photo, direction, equipment details | Contributed under ODbL-compatible terms (terms-of-service.md §3) | Pre-publication moderated; verified reports are upstreamed to OSM. |
| Vendor transparency portals, news, legislative databases | Context (retention, sharing, audit stats, votes, bills) | Various; restricted sources stay in a link-and-cite lane | These inform jurisdiction pages with per-claim citations; restricted-license material never enters the redistributable database. |
Every camera record carries per-source provenance: source type, URL, license, and retrieval date, visible on its dossier page.
2. Confidence tiers
Every camera record is in exactly one tier. Tiers are computed from the provenance record, never hand-assigned without a logged reason.
| Tier | Meaning | Requirements |
|---|---|---|
| UNVERIFIED | Imported from an external dataset (e.g., OSM) without independent confirmation by SFCN or its community, or a bare submission pending anything more | Valid coordinates + at least one recorded source |
| COMMUNITY_REPORTED | One credible first-hand report from an SFCN contributor, accepted through moderation | Report passing contributor-policy.md standards; photo strongly encouraged |
| COMMUNITY_VERIFIED | Two or more independent verifications — different contributors, no shared account signals, each attesting first-hand observation (or one report + one photographic verification by another party) | Independence checks per ../docs/MODERATION.md; verifier reputation weighting applies |
| OFFICIALLY_DOCUMENTED | Confirmed by a government or authoritative primary source: agency GIS layer, public-records response, contract/permit specifying the location, or agency-published map | Primary document archived and linked per source-standards.md |
Status flags (orthogonal to tiers, displayed prominently):
- DISPUTED — a good-faith challenge is open (wrong location, wrong equipment type, "not an ALPR," "never existed"). Shown with the dispute reason until resolved.
- REMOVED_CONFIRMED — credible evidence the camera has been taken down; kept in the historical record, excluded from active counts, shown in timelines.
- STALE — no source refresh, verification, or observation in 18 months. Stale records remain mapped (equipment usually persists) but are visually distinguished and prioritized for re-verification.
3. Deduplication rules
ALPRs cluster (multi-camera poles, both directions of a road), so dedupe is judgment plus rules:
- Candidate match: new reports within 25 meters of an existing record trigger a dupe check; 25–75 m triggers a soft warning to the submitter and a moderator flag.
- Distinct-camera signals that defeat a merge: different facing/direction of travel, different mount (pole vs. trailer vs. structure), different equipment type or manufacturer, photographic evidence of separate devices.
- Merges preserve everything: the surviving record inherits all sources, photos, verifications, and history from the merged record; the merged ID becomes a permanent redirect. Merges are moderator-only and logged (../docs/MODERATION.md).
- OSM reconciliation: records with an
osm_idsync against OSM edits; conflicts (moved/deleted upstream) route to the moderation queue rather than auto-resolving.
4. Jurisdiction assignment
- Every camera is assigned to a state, county (or county-equivalent), and where applicable city by point-in-polygon join against U.S. Census TIGER/Line boundaries (public domain), using county FIPS codes as the canonical key.
- Boundary edge cases (points on borders, enclaves, independent cities) resolve to the containing county-equivalent per TIGER; we display the assignment on the dossier so locals can dispute it.
- Population figures for per-capita statistics come from the most recent Census estimates, with vintage noted.
5. Rankings methodology
Our rankings (e.g., "most-documented counties") are computed as follows:
- Metric: count of active documented cameras (excludes REMOVED_CONFIRMED; includes all tiers, with tier mix disclosed) per jurisdiction, and per-capita variants per 10,000 residents.
- Tie-breaking: per-capita rank breaks ties on absolute count; small-population jurisdictions (<5,000 residents) are excluded from per-capita rankings to avoid absurd ratios, and this exclusion is stated wherever the ranking appears.
- Completeness caveats — displayed with every ranking, not buried here:
- Rankings measure documentation, not deployment. A county can rank low because it has few cameras or because no one has mapped it yet.
- Community coverage is uneven: documentation correlates with active local contributors, press attention, and open-data practices.
- We therefore say "most-documented," never "most-surveilled," unless the figure comes from official sources, and each ranked row shows its tier mix and data-recency indicator.
- Rankings never imply that an unranked or low-ranked jurisdiction is camera-free.
6. Update cadence
| What | Cadence |
|---|---|
| OSM extraction refresh (diff-based) | Weekly, with the snapshot date shown sitewide |
| Citizen reports & verifications | Continuous; published upon moderation approval |
| EFF Atlas re-import | On each Atlas release |
| Government-source re-checks | On alert triggers plus scheduled review |
| STALE sweep (18-month rule) | Monthly |
| Public ODbL dumps and diffs | Weekly, at [stopflockcamsnow.com/data] (see data-license.md) |
| Jurisdiction page statistics | Rebuilt on every data change (event cascade) |
Every data page displays its last-updated timestamp and the underlying snapshot dates.
7. Known limitations
- Coverage bias (§5 caveats) — documentation follows contributors.
- Equipment identification from photos can be wrong; manufacturer fields are best-effort, and AI-assisted identifications are always labeled AI ESTIMATE and never auto-verify (editorial-standards.md, ../docs/AI_RESEARCH.md).
- Operator attribution (which agency runs a camera) is harder than location; where inferred rather than documented, the dossier says so.
- OSM upstream data carries its own community's error modes; we inherit and, via verification, help fix them.
Found an error? See corrections-policy.md — we correct within 48 hours of confirmation.