Who is behind this

The Accurint Files is an independent investigation by Shawn Segal (pen name) — an investigative journalist documenting the intersection of private data brokers and public law enforcement. The project is editorially independent and reader-supported. It carries no advertising and no sponsored content.

Editorial principles

1. Show your sources. Every claim on this site links to a primary document, names a secondary source, or is explicitly flagged as an inference from those sources.
2. Distinguish what’s known from what’s suspected. The Confirmed / Strong Signal / Suspected hierarchy is published, not hidden. Readers see what kind of evidence backs each classification.
3. Honor the reader’s intelligence. A policy staffer and a curious neighbor get the same facts. Plain language is not the same as dumbed-down summary.
4. Empower the reader. Where action is possible — at a council meeting, in a FOIA request, in a question to a local official — we provide the templates and the language for the reader to take the action themselves.

How records are classified

Each agency record uses one of five Confirmation Status values. The classification rules:

• Confirmed Contributor — We hold an AVCC XML Addendum (or equivalent) containing the contribution-obligation clause AND the irrevocable-license clause. No inference required.
• Strong Signal — The Schedule A contains a Database Interface line item, or a consortium MOU names the agency as a member, but we don’t yet have the underlying addendum text.
• Suspected — The agency is named in advocacy reporting or procurement databases, or appears in a consortium’s member list, without independent contract confirmation.
• Reader Only — The agency has AVCC access (consumes data) but no evidence of contribution.
• Needs Investigation — Named in primary or secondary sources, but no contract evidence yet obtained. These records are not published until upgraded.

Publication gate

A record exists in our research database the moment it’s discovered. It does not appear publicly until a separate editorial review verifies the sourcing, writes a plain-English summary, and confirms no unresolved duplicates. The public site renders only records that have passed this gate.

Source priority

We prefer sources in this order:

1. Signed contracts and addendums
2. FOIA returns from agencies or via MuckRock
3. City/county council agenda packets and meeting minutes
4. Government procurement portals and SAM.gov
5. Reports from EFF, ACLU, EPIC, Brennan Center, and similar civil-liberties research organizations
6. Court filings referencing LexisNexis data
7. Investigative journalism by The Intercept, WIRED, 404 Media, and ProPublica
8. Academic research on police data sharing

We do not treat LexisNexis marketing materials, press releases, or product pages as evidence of any specific agency’s contribution. Those documents describe the platform’s capabilities, not who uses them.

Data Quality

Accuracy is foundational to this project. When an error or inconsistency is discovered in the database — whether introduced by human research, AI-assisted analysis, or data entry — we execute a structured remediation process to fix the immediate error and prevent systemic recurrence.

Human-sourced errors

When an error originates from human research or data entry, we:

1. Assess the nature and scope of the error
2. Retrain the researcher involved, with documented feedback
3. Implement preventive measures — improved checklists, additional review steps, or enhanced source-verification protocols — to ensure the error category doesn’t recur

AI-sourced errors

When an error is introduced or missed by an AI-assisted tool, we execute a deeper analysis to understand the root cause and fix it systematically:

1. Review the specific research instance. We retrieve the exact search prompt, the skill or agent configuration used during that research session, and any parameters that influenced the search.
2. Analyze the extraction logic. We trace the steps the agent took to identify and extract the incident or fact, identifying where the error branched from the correct path.
3. Develop a root cause analysis. We document why the error made it into the database — was it a prompt ambiguity, a flaw in how the agent prioritized sources, a misunderstanding of a contract clause, or a limitation in the knowledge base?
4. Fix systematically. The root cause determines the fix: we update the skill or agent logic, refine the knowledge base with clearer examples, revise the research prompts to eliminate ambiguity, or improve the extraction instructions.
5. Execute a retrospective remediation pass. Using the same configuration that led to the error, we audit all records researched during the affected period to identify similar errors or edge cases. This ensures that if the misconfiguration affected one record, we find and correct any other records from the same batch.

AI-assisted research

Some research is conducted with AI-assisted tools, including large-language-model summaries of dense contracts. AI-sourced claims are flagged in internal research notes as unverified until independently confirmed against a primary source. We do not publish AI-generated claims as fact, ever.

Corrections

We correct factual errors promptly and transparently. If a record on this site misstates a fact about your agency or jurisdiction, contact the tip line. Documented corrections will be reflected in the record with a note.

Tip line

Documents and tips are welcome. Particular interest in:

• Signed AVCC agreements or AVCC XML Addendums for any U.S. law enforcement agency
• Interface Control Documents (ICDs) or field-mapping schemas describing the RMS-to-LexisNexis upload
• Memoranda of Understanding for consortiums like CISC, CVISS, San Joaquin AVCC, or any other multi-agency data-sharing entity
• FOIA returns received from any agency, even partial returns
• Council agenda packets that include LexisNexis contract approvals

Contact: tips at segal.report (replace “at” with @). For sensitive documents, a Signal number and PGP key are available on request — email first without the document attached.

We do not publish sources without explicit consent. We do not run sourced material through cloud services that could log or retain it.