Glossary

Flockwatch relies on a vocabulary that mixes law-enforcement jargon, vendor product names, and computer-science abbreviations. The terms below are the ones that appear most often in the incidents registry, the legal landscape, and the vendor market context. Definitions lean on public sources; where the term is contested or used inconsistently across agencies, the entry says so.

When a definition is informed by a specific incident in the registry, it links to that incident so you can see the term used in context.

Reader's note

Two adjacent fields use overlapping vocabulary in inconsistent ways: law enforcement / CJIS-style usage (NCIC, ORI, BOLO) and computer vision / ML-style usage (OCR, mAP, model). This glossary distinguishes them where it matters — most entries are law-enforcement, but the model-evaluation terms used in the Error-Amplification Lab are flagged.

The glossary is intentionally conservative about scope: it defines what a general reader needs to understand the rest of the site, not every abbreviation that appears in a court filing. Vendor-specific product names (FlockOS, Axon Fusus, Motorola Vigilant) get entries only when they appear in the registry.

Core terms

ALPR (Automatic License Plate Reader)

A camera-and-software system that photographs a vehicle's license plate, runs optical character recognition (OCR) on the image, and compares the extracted text against a database of plates of interest. Also called ANPR (Automatic Number Plate Reader) in the UK and ANPR/ALPR interchangeably in vendor marketing.

ALPR systems are deployed in three forms: fixed (mounted on poles, signs, or street furniture at fixed locations), mobile (mounted on police vehicles, often with multiple cameras for rear/side coverage), and portable (battery-powered or temporary-mount units deployed for specific investigations).

Hot list (or hot-list)

A database of license plates the system is configured to alert on. Hot lists vary by source: NCIC (federal), state Bureau of Motor Vehicles (state), agency-specific BOLOs (Be On the Lookout), and shared commercial lists from vendors or fusion centers. A "hit" occurs when a scanned plate matches a hot-list entry.

The Flockwatch study documents multiple hot-list hygiene problems — most prominently in FW-002, where a rental vehicle remained on a stolen-vehicle list after it had been recovered.

BOLO (Be On the Lookout)

A specific alert placed on a plate or vehicle description by a law-enforcement agency. BOLOs are typically short-lived (hours to weeks) and agency-specific, though they may propagate to neighboring agencies through informal channels or commercial hot-list sharing. BOLOs are distinct from longer-lived warrants or stolen-vehicle flags.

NCIC (National Crime Information Center)

The FBI-maintained federal database that includes plates associated with wanted persons, stolen vehicles, missing persons, and similar categories. NCIC entries are governed by strict retention rules; an entry should be removed when the underlying case is resolved. NCIC access is regulated under 28 CFR Part 20 and audited by the FBI's CJIS Division.

Hit (or "hot hit")

A system notification that a scanned plate matches a hot-list entry. Hits require human confirmation before any enforcement action; the ALPR system itself does not authorize a stop. The hit rate (number of hits per day or per camera) is a key operational metric and a primary scaling concern in dense deployments.

ORI (Originating Agency Identifier)

A unique nine-character identifier assigned by the FBI to each law-enforcement agency. ORIs are the unit of access control and audit logging in NCIC and state CCH systems. Audit logs at the agency level are typically indexed by ORI.

Vendor and product terms

FlockOS

The software platform that operates Flock Safety's camera network. Includes the OCR pipeline, cross-agency search and sharing settings, and the audit log. (Falcon is Flock's fixed LPR camera line, not the search interface.) Flock's sharing model offers agency, statewide, and nationwide lookup scopes; Flockwatch has found no public document establishing which scope is applied by default, and treats the effective breadth of access as an open question — see FW-005, FW-006, and FW-019, where a nationwide setting was applied without the city's knowledge.

Flock Safety

The largest commercial vendor of networked ALPR systems in the United States, with deployments in approximately 5,000+ communities as of 2026. Founded 2017; headquartered in Atlanta, GA. The company's cameras are primarily fixed-mount and networked through FlockOS.

Axon Fusus

Axon's ALPR-adjacent product, integrated into the broader Axon Evidence / Axon Network suite. Fusus focuses on real-time crime-center aggregation (camera streams + license plate reads + gunshot detection) rather than direct ALPR deployment.

Motorola Vigilant

A competing ALPR vendor focused on mobile (vehicle-mounted) cameras. Notably used by the DEA and other federal agencies in addition to local police. Vigilant's commercial hot-list product is sometimes marketed as a supplement to NCIC for non-law-enforcement commercial customers (e.g., repossession agencies, hotel security).

BriefCam

Video analytics vendor (acquired by Canon Inc. in 2018) that integrates with ALPR systems to provide object detection, search, and summarization across recorded footage. Not an ALPR vendor directly but commonly bundled in surveillance-platform procurements.

The practice of making a local ALPR hot-list searchable to other agencies, either by direct network connection or via a vendor's shared platform. The Flockwatch study documents cases where this sharing reached agencies the camera owner had not authorized — most prominently FW-019 and FW-020 — and cases where a purpose limit failed despite the sharing itself being permitted (FW-005).

Operational and policy terms

Purpose limitation

A principle in information-governance law and policy that data collected for one purpose (e.g., locating a stolen vehicle) may not be used for a substantially different purpose (e.g., tracking attendance at a political protest) without additional legal authority. Purpose limitation appears in the legal landscape section as a recurring gap between vendor policy and statutory restriction.

Audit log

A persistent record of every query, search, alert, and access to ALPR data. Audit logs are critical for post-hoc investigation of misuse (FW-010, FW-011). There is no federal ALPR audit-logging requirement; among the ten jurisdictions surveyed on the regulation page, only some (e.g. Vermont, North Carolina) require audits by statute. Where audit logging exists it is usually a vendor feature or an agency policy rather than a statutory mandate — which is why the breakpoints treat independent audit as a control that must be imposed, not assumed.

Retention period

The length of time ALPR data (read events, plate images, search history) is stored before automatic deletion. Retention periods are the most variable element of ALPR policy — some agencies retain 30 days, others retain indefinitely. The Flockwatch study documents cases where retention period claims in public-facing policy documents diverge from actual practice.

Civil immigration enforcement access

The question of whether local ALPR data is accessible to federal immigration authorities (ICE, CBP). This is the single most-contested operational question in ALPR policy as of 2026 and is the trigger for many of the cancellations documented in community pushback.

Sanctuary policy / sanctuary jurisdiction

A municipal or state policy limiting local cooperation with federal civil immigration enforcement. Sanctuary policies vary widely in scope and legal force. The Flockwatch study does not take a position on sanctuary policy as such but documents several cases where the discovery of ALPR data access by immigration authorities motivated contract cancellation despite a sanctuary policy being nominally in force.

Data and technical terms

OCR (Optical Character Recognition)

The computer-vision step that converts a plate image into text. OCR accuracy on license plates depends on plate design (font, contrast, state), environmental conditions (lighting, weather, angle), and camera quality. The Error-Amplification Lab uses an open OCR model to study how OCR errors propagate through the ALPR pipeline.

Hit rate

The number of hot-list matches per unit time or per camera. A high hit rate is generally desirable from an operational standpoint (more potential apprehensions per camera-day) but can also indicate a poorly-curated hot list, as documented in FW-002.

Plate crop

The bounding box of the license plate within a larger camera image, extracted before OCR. Plate crops may be stored independently of the original image in some vendor systems. The Error-Amplification Lab generates plate crops as part of its deterministic string-collision experiment.

String collision (or plate collision)

A pair of distinct plate strings that produce the same (or nearly the same) OCR output due to character-substitution error (e.g., 8 vs B, 0 vs O, I vs 1). Substitution collisions point at the wrong single candidate; deletion collisions — the mechanism the Error-Amplification Lab reproduces — collapse many distinct identifiers onto one key. The lab demonstrates the cardinality of that ambiguity set (100 identifiers → 1 key), not a collision rate; no public data supports a rate estimate.

mAP (mean Average Precision)

An object-detection metric from computer vision, common in ALPR detection benchmarks. It is defined here because vendor and academic ALPR papers report it. Flockwatch does not compute mAP. The lab reports the exact_matches and middle_token_preserved counts across six degradation conditions on a single open font; see the Error-Amplification Lab.

Confidence score

A numeric output from the OCR model indicating the model's certainty that the recognized text is correct. Confidence scores are sometimes used to suppress low-certainty reads before they enter the hot-list comparison; the policy varies by vendor. Flockwatch does not publish a confidence-calibration analysis: the open model's confidence outputs are requested but not retained in the 600 published inference rows. Whether an operational ALPR surfaces confidence to the officer is governance breakpoint 1, not a measured result here.

Adversarial input

A deliberately-crafted image designed to cause OCR errors. Most commonly used in security research to evaluate OCR robustness. Outside security research, adversarial inputs occasionally appear in operational contexts (e.g., obscuring plates to defeat ALPR); some states have laws against this.

Legal and regulatory terms

Statutory authority

The specific statute or ordinance that authorizes a government agency to deploy ALPR systems. Statutory authority varies by jurisdiction and may be express (a state ALPR-specific law), implicit (a general law-enforcement authority), or unclear. The legal landscape section documents the variability.

CJIS (Criminal Justice Information Services)

The FBI division that regulates access to federal criminal-justice databases including NCIC. CJIS Security Policy governs how state and local agencies handle CJIS data and includes specific requirements for ALPR systems that integrate with NCIC.

28 CFR Part 20

The federal regulation governing criminal history records and the operation of state criminal-records repositories. Includes provisions on accuracy, completeness, and timeliness that bear on NCIC entry hygiene.

State Public Records Act

The state-specific statute governing public access to government records. Often the legal mechanism by which journalists and researchers obtain ALPR-related documents (audit logs, vendor contracts, retention statistics). The strength of state Public Records Acts varies significantly.

Qualified immunity

A legal doctrine that shields government officials from civil liability for actions taken in the course of their duties, unless those actions violate clearly established statutory or constitutional rights. Qualified immunity is a recurring issue in ALPR-related civil-rights litigation, including several cases in the incidents registry.

Section 1983 (42 U.S.C. § 1983)

The federal civil-rights statute under which most ALPR-related constitutional claims are brought (typically as Fourth Amendment unreasonable-search claims). Section 1983 actions are subject to qualified-immunity defenses.

Reader-facing terms

Evidence grade

A Flockwatch-specific classification of how well-sourced an incident claim is. The two grades used in this site:

Grade A — primary or official source plus corroboration (court opinion, government release, audit, settlement record, or linked public report).
Grade B — strong multi-source journalism or public-records analysis; the complete primary record is not in hand.

Pending complaints and criminal charges remain allegations until adjudicated.

FW-### (case identifier)

The Flockwatch case identifier used throughout the site. FW-001 through FW-029 are the cases currently in the registry. The registry table and incidents.csv are the canonical record for each case; references to FW-### elsewhere on the site point back to that row.

Hermanity

The family of sites that share the canonical hermanity Hugo theme. Sister sites include blue-lotus.hermanity.dev (field brief), hermanity.dev (personal site), and others listed at the sites directory.

Contested and inconsistent terms

"Real-time"

Vendor marketing and agency policy sometimes differ on what "real-time" means for ALPR. The narrowest definition: plate text is available for hot-list comparison within ~1 second of the camera reading the plate. The looser definition: plate text is available within a few minutes, batched into searchable history. The Flockwatch study uses the narrow definition unless otherwise noted.

"Anonymized" / "de-identified"

ALPR systems store plate reads as text (the plate string itself), not as anonymized hashes. True anonymization is not standard in commercial ALPR deployments. Some vendor policies describe short-retention aggregation as "anonymization" in a way that does not match the technical definition.

"Network"

"Flock Network" or "the network" in Flock Safety marketing refers to the cross-agency sharing layer of FlockOS. Public materials describe agency-selected sharing scopes, including specific-agency, statewide, radius-based, nationwide, and no outside sharing. Flockwatch treats the scope actually configured for a camera owner as a fact to establish from records, not as a universal default.

"Audit"

Vendor "audit" can mean (a) the system's internal access log (which the vendor controls), (b) a third-party audit of vendor or agency practices, or (c) an oversight body's review. The Flockwatch study distinguishes these explicitly.

What's not in this glossary

A few categories of terms are intentionally excluded:

Court-specific procedural terms (motion in limine, summary judgment, etc.) — these are general legal vocabulary, not ALPR-specific.
Vendor-internal engineering terms (specific API endpoints, internal product names) — these are not in the public record.
Specific state-law abbreviations — these vary by jurisdiction; the legal landscape section covers the high-level landscape.
Computer-vision research terms beyond mAP — readers wanting the deeper CV vocabulary should consult the academic literature cited in the evidence base.

If a term you encountered on this site is missing, please open an issue at the GitLab project.

Last reviewed 2026-08-20, the date of the most recent registry update. Flockwatch does not provide legal advice; this glossary is descriptive, not normative.