FIELD STUDY 01 · COMPUTATIONAL CRIMINOLOGY · JUL 2026

When uncertain data becomes police force.

A public-data study of how camera reads, hot lists, national sharing, and human verification failures combine inside networked automated license-plate-reader systems.

Read the full paper Reproduce the analysis
No private Flock access · no live vehicle queries · no UCR proxy
01 / WHAT THE DATA SUPPORT

Three findings. Three different denominators.

FOOTPRINTDOCUMENTED
113,963

publicly mapped cameras

Fifty-state OpenStreetMap-derived census. A lower bound—not a vendor inventory.

CAPTUREMASS SCALE
99.890%

without a contemporaneous hit

California 2021–22 detections. “Non-hit” does not mean innocent or irrelevant.

FAULT LABREPRODUCED
100:1

identifier compression

Transparent synthetic middle-token-loss model; not a Flock accuracy estimate.

02 / EVIDENCE CORPUS

Scale, activity, cases, and local tests.

  • 4,084Atlas ALPR recordsagency / technology rows
  • 2,629records naming Flock64.4% of Atlas ALPR rows
  • 3.224BCalifornia detections2021–22 public records
  • 3.557Mrecorded hot-list hits0.110% of detections
  • 13documented incidentsevidence-graded, not prevalence
  • 600synthetic OCR inferencesopen model · local CPU
03 / WORKING CONCLUSION

The decisive error is often not the first one.

A bad read becomes dangerous when interfaces hide ambiguity, databases distribute it, policies permit broad access, and officers treat the alert as self-authenticating suspicion. Governance has to break the whole chain.

Review twelve breakpoints →
04 / MEASURED OUTPUTS

Five public datasets. One synthetic mechanism lab.

Horizontal bar chart ranking states by publicly mapped ALPR cameras per 100,000 residents01 Camera density Horizontal bar chart of EFF Atlas ALPR records by named vendor02 Vendor records Percentage chart showing California ALPR detections recorded as hits versus no hit03 Detection burden Grouped bar chart showing exact string recovery and middle-token preservation across six synthetic OCR conditions04 OCR degradation
05 / EXPLORE THE RECORD

From deployment to remedy.

01DEPLOYMENTNational footprint→ 02MECHANISMError-amplification lab→ 03CONSEQUENCEIncident registry→ 04DOCTRINELegal landscape→ 05VALIDITYMethods and limits→ 06REPLICATIONData and code→