Appendix B — Reference Library

This appendix is the manual’s reference library. The body of the manual explains how Actual Vote works conceptually, methodologically, operationally, and philosophically. The reference library provides supporting material that is too detailed, too specialized, or too document-like to belong in the main flow.

There are sixteen reference sections. Some are reference material authored for this manual; some are reproduced canonical documents; some are stubs for reference works that will be developed over time.

B.1 — Who’s Who

This appendix identifies the people and organizations mentioned in the manual, with brief descriptions and their relevance to the Actual Vote ecosystem.

People

Dan Wolf. Founder and CEO of America Counts (Democracy Counts, Inc.). Background: Harvard Law, UCSD political science, Nicaragua election observation (1984), TransBorder Institute, landmine detection technology companies. Recognized that mobile technology could enable citizen-powered election audits after studying the 2004 Ohio election and historical fraud cases. Responsible for institutional design, legal framework, fundraising, and strategic partnerships. First mentioned: Section 2.5.

Jason Flatley. CTO and methodology architect of America Counts. Responsible for the Actual Vote methodology, data science, app development oversight, analysis operations, and field operations. The primary technical authority for the Comparison Analysis pipeline. Contact: j.flatley@democracycounts.org. First mentioned: Section 1.1.

Bennie Smith. Election integrity advocate in Shelby County, Tennessee. His documentation of unreported votes from predominantly Black precincts in the 2015 Shelby County election is one of the canonical examples of the kind of vote reporting failure that AV is designed to detect. See Shelby County 2015.

Marilyn Marks. Executive director of the Coalition for Good Governance, a non-profit focused on election integrity litigation and advocacy, particularly regarding electronic voting systems. A partner and ally in the broader election integrity landscape.

Lynn Bernstein. Civic organizer and election integrity advocate. Involved in the Citizens Audit Broward (CAB) effort.

Eric Olsen. Election integrity advocate and partner organization leader. Involved in coordination with AC field operations.

Organizations

Democracy Counts, Inc. The 501(c)(3) registered non-profit corporation that is the legal parent entity of America Counts. Founded 2016 by Dan Wolf. Corporate website: democracycounts.org. Donations are tax-deductible to the extent allowed by law. First mentioned: Section 2.5.

America Counts (AC). The public-facing name for the non-profit non-partisan election audit initiative operated by Democracy Counts, Inc. Operates the Actual Vote system, publishes analyses, and advocates for independent vote reporting verification. Website: americacounts.us. First mentioned: Section 1.1.

Citizens Audit Broward (CAB). A local civic group in Broward County, Florida, that operated as an audit committee in 2018 and 2020, conducting Comparison Analyses of Broward County elections. The canonical example of an audit committee in this manual. First mentioned: Section 7.4.

Scrutineers. A partner organization that helps connect volunteers with AC and supports poll tape recording efforts across multiple jurisdictions. Provides organizational infrastructure for field operations. First mentioned: Section 2.5.

Verified Voting. A non-profit organization focused on voting equipment standards, research, and advocacy for verifiable elections. Maintains a comprehensive database of voting equipment in use across the United States (verifiedvoting.org). Complementary to AV’s operational focus. First mentioned: Section 13.5.

Election Assistance Commission (EAC). The federal agency that assists state and local election officials in the administration of federal elections. Develops voluntary voting system guidelines and maintains the national voter registration form. Relevant to AC’s recommendations regarding federal standards for vote reporting transparency. First mentioned: Section 15.2.

National Conference of State Legislatures (NCSL). A bipartisan organization serving state legislators and staff. Maintains comprehensive resources on state election laws, including poll watcher and observer rules, which AC references for state-specific guidance. First mentioned: Section 6.

State Voices. A national network of civic engagement organizations working on voter registration, civic participation, and election integrity. A potential partner in the broader election integrity landscape. First mentioned: Section 13.5.

B.2 — Index of AC Analyses to Date

This appendix provides a master index of all Actual Vote analyses conducted by America Counts, organized chronologically. It is updated with each new analysis cycle.

Master table

Notes on the table. Dashes indicate values not yet compiled into the master index. The Broward 2018 operation focused on documenting posting compliance rather than conducting full Comparison Analyses. The Broward 2020 analyses were conducted by Citizens Audit Broward (CAB) and their publication status is pending. The NC 2024 Primary is the canonical complete analysis. The 2024 General Election analyses are documented in the 2024 General Election News Page and their final accounting is ongoing as of this manual’s publication.

This table will be updated as new analyses are completed and as historical analyses are compiled into the standardized format.

B.3 — State-by-State Reference

This appendix is a reference catalog for each US state where AC has operated or researched. For each state, the full version provides the information an Actual Vote user or analyst needs to operate effectively in that jurisdiction.

Framework

For each state, the full reference includes: an overview of the state’s election administration structure; poll tape posting practices and legal requirements; viability of each access method (Methods 1–5); voting method categories used in official results; comparison grain notes (whether the standard County × Precinct × Voting Method × Contest × Choice grain works or requires adaptation); per-county notes for counties where AC has worked; reporting cadence and certification timeline; past AC analyses; and open questions.

North Carolina

AC’s most-worked-in state. Multiple Comparison Analyses, including the canonical NC 2024 Primary analysis and the 2024 General Election rolling-log analysis. Counties with significant work include Wake, Mecklenburg, Henderson, McDowell, Guilford, Forsyth, and others.

Election administration is handled by county Boards of Elections under the oversight of the NC State Board of Elections. NC requires poll tape posting after polls close — Method 1 is viable statewide. Elections offices in NC vary in cooperation level but the general pattern supports Method 3 (office visits). NC Public Records Law applies for Method 4 (FOIA) requests. Voting machine technology varies by county. The standard comparison grain works for NC Election Day voting; Early Voting may require adaptation depending on county reporting structure.

NC official results are typically published at the precinct × voting method level through the NC State Board of Elections website. Certification occurs on a county-by-county basis followed by state canvass. The analysis window timing is documented in the NC 2024 Primary Report (B.6).

Georgia

Notable for the vote-center grain example documented in Section 8.6. Georgia reports Early Voting at the vote center level rather than the precinct level, which requires comparison logic adaptation. The 2024 General Election analysis included 326 Georgia comparisons.

Election administration is handled by county Boards of Elections and Registration under the oversight of the Georgia Secretary of State. Georgia poll tape posting practices vary by county. The standard comparison grain requires modification for Early Voting (vote center substituted for precinct).

Florida

Significant AC experience, particularly through the Citizens Audit Broward (CAB) effort in 2018 and 2020 and follow-on Florida work in the 2024 General Election. Elections are administered at the county level by Supervisors of Elections (SOEs).

Florida poll tape posting practices vary by county; Method 1 viability depends on the county. Broward County specifically was the focus of the Broward 2018 posting compliance documentation, which found approximately one-third posting non-compliance even with clear legal requirements. Florida Sunshine Law applies for Method 4 requests. The 2024 General Election analysis included 399 Broward County comparisons.

Texas

Known for Method 2 viability under the poll watcher framework. Texas Election Code Chapter 33 permits poll watchers appointed by candidates, political parties, and certain other entities to observe election proceedings, including the printing and handling of poll tapes. Texas voting machines typically print multiple copies of results tapes, of which one can be posted publicly (Method 1) and others can be observed or recorded by poll watchers (Method 2). Texas has not yet been the focus of a published full analysis but is a high-priority target for future operations.

Other states

AC has conducted preliminary research on election administration practices in Virginia, New Jersey, Tennessee, Ohio, Illinois, Wisconsin, and Arizona. These states are candidates for future operations and will be documented in this appendix as AC develops operational experience in each. The framework above establishes the structure for these entries.

Template for unpopulated states

For states not yet documented, the following template should be completed as AC develops experience:

1. Overview of election administration structure
2. Poll tape posting practices (statute citations)
3. Method 1–5 viability assessment
4. Voting method categories
5. Comparison grain notes
6. Per-county notes (as applicable)
7. Reporting cadence and certification timeline
8. Past AC analyses
9. Known contacts or precedents
10. Open questions

B.4 — User Agreement

Status: Reproduced document.

The Actual Vote User Agreement is the canonical AC document that governs the relationship between Actual Vote users and AC. It is reproduced in this appendix in full, exactly as it appears on the AC platforms at the time of v2.0 publication. Cross-referenced from Section 7.7 (In-App Consent Flow).

The User Agreement covers: the user’s grant of rights to AC for the use of submitted recordings; AC’s commitments to the user regarding the handling of recordings (including the vetting process and rejection criteria); the user’s responsibilities (including respecting property owner instructions, not capturing PII of others where avoidable, and not using the app in ways that would expose the user or others to harm); AC’s no-guarantees position on outcomes (analyses are not guaranteed to be conducted or published; findings are not guaranteed to be acted upon by elections offices or courts); changes to the agreement (how AC will notify users of changes); and termination (how either party can end the relationship).

The current User Agreement is available on the AC website and within the Actual Vote app. Rather than reproducing it here (where it would risk becoming outdated as the agreement is revised), this appendix describes its contents and directs readers to the canonical source.

Canonical source: The current User Agreement is published at americacounts.us and is presented to users during the in-app consent flow described in Section 7.7. The version in effect at the time of v2.0 publication governs; subsequent revisions will be noted in the AC Findings Changelog (B.8).

B.5 — Privacy Policy

Status: Reproduced document.

The Actual Vote Privacy Policy is the canonical AC document that describes how AC handles user data. It is reproduced in this appendix in full, exactly as it appears on the AC platforms at the time of v2.0 publication. Cross-referenced from Section 7.7 (In-App Consent Flow).

The Privacy Policy covers: what data AC collects from users (account information, submission metadata, recording content); what data AC does not collect (voting history, party affiliation, demographic information, political views); how AC uses the data (for analysis purposes, for chain-of-custody, for communication with users); how AC shares the data (with whom and under what conditions); how AC protects the data (security measures, retention policies); user rights (access, correction, deletion of personal data); cookies and tracking; children’s privacy; and changes to the policy.

The current Privacy Policy is available on the AC website and within the Actual Vote app. Rather than reproducing it here, this appendix describes its contents and directs readers to the canonical source.

Canonical source: The current Privacy Policy is published at americacounts.us and is presented to users during the in-app consent flow described in Section 7.7. The version in effect at the time of v2.0 publication governs; subsequent revisions will be noted in the AC Findings Changelog (B.8).

B.6 — NC 2024 Primary Election Report

Status: Reproduced document.

The NC 2024 Primary Election Report is the canonical published example of an Actual Vote analysis report. It documents the analysis of the March 5, 2024 North Carolina primary election across seven counties, with 52 transcribable submissions, 4,625 transcribed values, 321,502 votes audited, and zero apparent discrepancies in the County/Precinct/Voting Method triples. Published in the before-certification window, giving it both operational relevance and archival value.

The report is reproduced in this appendix in full, exactly as it appears on the AC website at the time of v2.0 publication. Cross-referenced throughout the manual as the exemplar of a complete AV analysis, particularly in Section 8 (Methodology) and Section 12.1 (Report Format and Structure).

Canonical source: The NC 2024 Primary Election Report is published on americacounts.us. Readers should consult the published version for the complete text, including the full comparison tables and investigation notes. The report is the single best example of what a complete Actual Vote analysis looks like in practice.

B.7 — 2024 General Election News Page

Status: Reproduced document.

The 2024 General Election News Page is the canonical published example of a rolling operational log format (Section 12.1). It covers a multi-state operation (North Carolina, Georgia, Florida) during the November 2024 general election and its aftermath, with entries documenting findings as they were produced, operational challenges as they arose, and investigation updates as apparent discrepancies were resolved.

The news page is reproduced in this appendix in full or substantially excerpted, exactly as it appeared on the AC website. Cross-referenced in Section 10 (Operational Practice) and Section 12.1 (Report Format and Structure).

Canonical source: The 2024 General Election News Page is published on americacounts.us. Readers should consult the published version for the complete rolling log, including timestamped entries documenting the multi-state operation as it unfolded. The news page is the single best example of what a rolling operational log looks like in practice.

B.8 — AC Findings Changelog

This appendix tracks substantive updates to AC’s findings, methodology, and recommendations. It is distinct from the version log (B.13), which tracks changes to the manual itself. The findings changelog tracks changes to the underlying work that the manual documents.

Format

Each entry records: the date of the change, what changed, why it changed, and which sections of the manual are affected.

Entries

v2.0 publication (current). Initial changelog entry. The findings, methodology, and recommendations documented in this manual reflect the state of AC’s work as of the v2.0 publication date. This changelog will be updated as new analyses are completed, as existing analyses are updated with new information, as the methodology evolves, and as recommendations are revised based on new experience.

Future entries will be added here as substantive changes occur. The changelog provides a transparent record of the evolution of AC’s work, complementing the version log’s record of the evolution of this manual.

B.9 — Bibliography and Recommended Reading

This appendix consolidates the sources cited throughout the manual and provides recommended reading for readers who want to go deeper.

AC canonical documents

These are documents created by America Counts that the manual treats as canonical sources.

The NC 2024 Primary Election Report — the canonical example of an Actual Vote analysis report. Reproduced in B.6. The 2024 General Election News Page — the canonical example of a rolling-log analysis. Reproduced in B.7. The Background on Actual Vote for Software Development document — internal AC document describing the system architecture. The Actual Vote Data Standards document — internal AC document specifying the data structures used in the analysis pipeline. The Privacy Policy — reproduced in B.5. The User Agreement — reproduced in B.4. Using Actual Vote is Legal — AC’s document on the legal basis for Actual Vote. Why You Can Trust Actual Vote — AC’s essay on the trust posture. Safety and Privacy When Using Actual Vote — AC’s document on user safety. About America Counts — AC’s institutional self-description.

AC technical papers

Three technical papers developed by AC provide the theoretical foundations for aspects of the work documented in this manual. Pointers to these papers appear in Section 13.5.

The Stochastic Harm paper develops a framework for understanding how election integrity violations propagate indirect harm through the political system. The Social Well-Being (SWB) paper develops a framework for quantifying the collective benefit of transparent elections in dollar-equivalent units. The Election Insurability paper frames election integrity as an insurable risk, asking whether a rational insurer would underwrite a policy against election integrity failure.

External legal and reference sources

Bert P. Krages II, The Photographer’s Right (2016) — the canonical statement of US photography law that AC relies on for the legal basis of Actual Vote, cited in Section 6.1. Texas Election Code, Chapter 33 — the legal authority for Texas poll watcher rules, cited in Section 6 and Section 7.3. Florida Sunshine Law — state public records law, cited in Section 7.5. NC Public Records Law — state public records law, cited in Section 7.5. Texas Public Information Act — state public records law, cited in Section 7.5. California Public Records Act — state public records law, cited in Section 7.5. Wisconsin Statutes — state law for the election inspector framework, cited in Section 7.3.

Recommended reading on elections

The following books are recommended for readers who want deeper context on the issues that AV addresses. This list draws from the curated reading list on the America Counts website.

Votescam: The Stealing of America by James and Kenneth Collier — foundational investigative work on vote fraud in American elections. Code Red: Computerized Elections and the War on American Democracy by Jonathan Simon — analysis of electronic voting vulnerabilities. Democracy Betrayed: The Wilmington Race Riot of 1898 and Its Legacy — historical treatment of political violence and vote suppression. Ballot Battles: The History of Disputed Elections in the United States by Edward Foley — legal history of contested elections, relevant to understanding the legal pathways in Section 6.5.

Additional titles from the AC recommended reading list should be added as the list is maintained and updated.

External web sources

The following external sources were consulted during the manual’s preparation: Wikipedia entries on election administration terminology; National Conference of State Legislatures (NCSL) pages on election worker laws and observer rules; state-specific election administration websites (Virginia Department of Elections, Texas Secretary of State, Wisconsin Elections Commission); U.S. Election Assistance Commission (EAC) resources; Verified Voting Foundation database (verifiedvoting.org); and Bipartisan Policy Center publications on poll watcher roles.

Published articles about America Counts

Published media coverage and articles about America Counts should be cited here with full citations and links. This section will be populated as AC compiles its media coverage record.

B.10 — Technical Data Standards Reference

This appendix summarizes AC’s internal data standards for structuring analyses. It is useful for anyone conducting, reviewing, or building upon an Actual Vote analysis.

Unique ID format

Each Actual Vote analysis has a unique identifier of the form [date]_[country]_[state]_[scope]. For example, 20240305_usa_nc_multi identifies the analysis of the March 5, 2024 election in multiple North Carolina counties. The date is in YYYYMMDD format. The country code is the ISO 3166-1 alpha-3 code (always usa for current analyses). The state is the two-letter postal abbreviation in lowercase. The scope is either a county name (lowercase) for single-county analyses or multi for multi-county analyses.

Folder structure

The current analysis pipeline stores working materials for each analysis in a structured folder. The standard structure:

Actual Vote analyses/
  [analysis_unique_id]/
    01_submissions/        — Inventory of submissions for this analysis
    02_official_results/   — Downloaded official results files
    03_transcription/      — Transcription spreadsheets
    04_comparison_logic/   — Mapping tables, aggregation rules, edge cases
    05_comparison_output/  — Results of running the comparison
    06_investigations/     — Investigation notes and outcomes
    07_context/            — Context submission inventories
    08_report/             — Draft and final report files
    09_correspondence/     — Communication with elections offices

The structure is informal in the sense that there is no enforced schema — different analyses may organize their files slightly differently. The organization is flexible enough to accommodate analyses of different sizes and complexity, but consistent enough that someone familiar with the pattern can navigate any analysis’s files.

CSV field names

The standard field names for comparison output files:

Discrepancy code list

The discrepancy code list categorizes why a non-match arose or how it resolved, aligned with the comparison-outcome model (Section 4.5 and Section 12.3). The codes fall in three groups.

The Coverage / access codes describe absence of data rather than a vote-reporting discrepancy. This list replaces the earlier six-code list, which predated the comparison-outcome model; the cumulative metadataset’s disposition categories (the B.10 funnel table) are being brought into line with these codes as part of the data-standards update.

Contest and candidate mapping

When poll tape contest and candidate names do not match official results labels exactly, a mapping table is required. The mapping table for each analysis is stored in the comparison logic folder and documents the exact correspondence between poll tape labels and official results labels. Naming inconsistencies are one of the most common sources of comparison friction (Section 15.1, Recommendation 8).

Cumulative metadataset

Alongside the per-analysis standards above, AC maintains a cumulative metadataset: a derived, queryable index covering every packaged analysis, regenerated mechanically from each package’s manifest and published summary — never edited by hand. It has two tables. The analyses table holds one row per analysis: identity (the unique ID), election metadata (label, date, state, counties, certification date, analysis window), report metadata (report ID, site URL), and the normalized headline statistics — transcribable submissions, transcribed values, comparisons, exact matches, explained matches, apparent discrepancies, unresolved discrepancies, and votes audited. Two identities are enforced when the metadataset is built: comparisons = exact matches + explained matches + apparent discrepancies, and transcribed values = comparisons + values with no comparison. (The category names here — explained matches, unresolved discrepancies — predate the comparison-outcome model; the metadataset schema and these identities will be reconciled with the new vocabulary as part of the data-standards update noted in Section 4.5.) The funnel table holds one row per analysis × disposition category — the discrepancy dispositions across analyses, which is what makes the trend analysis contemplated by the discrepancy code list (Section 4) actually queryable. The cumulative metadataset is the source of truth for cross-analysis running totals; the published dashboard’s numbers are checked against it.

B.11 — Election Administration Terminology Reference

This appendix consolidates the election administration terminology used throughout the manual. Election administration in the United States uses different terminology in different states: the same role may be called by several different names, and the same office may be structured differently. A reader whose local elections office is called the “Registrar of Voters” may be confused when the manual refers to “the elections office” or “the SOE.” This appendix maps the variations.

Generic terms used in the manual

This manual uses several generic terms that correspond to different state-specific titles. Elections office is the generic term for the government office that administers elections in a jurisdiction. Poll tape is the strip of paper printed by a voting machine at the close of polls, showing the totals tabulated by that machine; the underlying object is similar across machine types but the print format varies. Polling place is the physical location where voters go to vote in person. Precinct is the smallest administrative unit at which votes are typically reported, though the actual structure varies widely (some precincts are single physical locations, some are geographic units that vote at shared physical locations, some are merged or consolidated for specific elections). Poll worker is a person formally appointed to staff a polling place during elections. Poll observer is a person formally appointed to observe election proceedings at a polling place.

State-specific terminology variations

The following variations are documented based on AC’s operational experience and research.

Elections office: In Florida, the Supervisor of Elections (county-level, abbreviated SOE). In North Carolina, the Board of Elections (state level: NC State Board of Elections; county level: County Board of Elections, often abbreviated BOE). In California, the Registrar of Voters (county level). In Wisconsin, the Municipal Clerk for many functions, with the Wisconsin Elections Commission at state level. In Virginia, the Department of Elections at state level and the General Registrar and Electoral Board at local level. In many other states, the County Clerk is the generic equivalent.

Poll worker: In Wisconsin, Election Inspector (must be a qualified elector of the county and must complete training). In Virginia, Officer of Election (appointed by Electoral Boards from political party nominations, serving up to three-year terms). In Ohio, Precinct Election Official. In many states, poll worker or election judge (generic).

Poll observer: In Texas, poll watcher (appointed by candidates, political parties, and certain other entities, governed by Texas Election Code Chapter 33). In Florida, poll watcher (appointed by candidates, political parties, or political committees, with each watcher requiring approval from the county SOE). In Colorado, election watcher (appointed by candidates, political parties, and issue committees; must receive a certificate of appointment and take an oath). In many states, election observer or poll watcher (generic).

Categories of voting methods

The voting method labels used throughout the manual and in AC’s comparison logic: Election Day refers to votes cast in person at the voter’s polling place on the official election day — historically the most common voting method, though declining as a share of total votes as other methods grow. Early Voting refers to votes cast in person at a designated location before the official election day, available in most US jurisdictions with varying time windows and location structures. Absentee refers to votes cast by mail in advance of the election — sometimes used interchangeably with “Mail-In Voting” but technically distinct in some jurisdictions, where “Absentee” historically referred to votes requiring a specified excuse. Mail-In Voting is the broader term for any vote cast by mail, including no-excuse absentee and universal vote-by-mail; some states use this as the primary or only voting method. Provisional refers to votes cast at a polling place by a voter whose eligibility is uncertain, counted separately and only after eligibility is verified. Other covers various edge categories including military and overseas ballots, federal write-in ballots, and ranked-choice ballots in jurisdictions that use ranked-choice voting.

Reporting structures

The structure of officially reported results varies by state and jurisdiction, which directly affects the comparison grain that AC uses for each analysis (Section 4.5, Section 8.6). The typical reporting grain is Precinct × Voting Method × Contest × Choice, which most jurisdictions use for in-person voting. Some jurisdictions report only at County × Voting Method × Contest × Choice for some voting methods (commonly absentee or mail). Some jurisdictions use Vote Center × Voting Method × Contest × Choice for Early Voting (notably Georgia), where Early Voting is administered at the vote center level rather than the precinct level. Combined precincts occur when two or more precincts vote at a shared physical location and may have results reported jointly or separately depending on state policy. Anonymity-redacted precincts occur when jurisdictions with very small numbers of voters redact precinct-level results to preserve voter anonymity; AC’s comparison grain must omit these as well.

B.12 — FOIA / Public Records Request Templates

This appendix provides template language for submitting public records requests for poll tape images or data, as described in Section 7.5 (Method 4).

Template request letter

The following template can be adapted for the specific state’s public records law and the specific elections office:

[Date]

[Name of Records Custodian or Elections Office] [Address]

Re: Public Records Request — Poll Tape Results Tapes

Dear [Records Custodian / Supervisor of Elections / County Clerk]:

Pursuant to [State Public Records Law — e.g., “the North Carolina Public Records Law, N.C.G.S. § 132-1 et seq.” or “the Florida Public Records Act, Fla. Stat. § 119.01 et seq.”], I am requesting copies of the following records:

Results tapes (poll tapes) printed by voting machines at the close of polls for the [election name and date] election in [county/jurisdiction name], for all precincts and all voting methods.

I am requesting these records in the form of scanned images (PDF or JPEG) or physical copies, whichever is more readily available and cost-effective.

If scanned images are available, they may be provided electronically to [email address].

If there are fees associated with this request, please advise me of the cost before processing. I am willing to pay reasonable copying and processing fees as provided by law.

If any portion of this request is denied, please cite the specific statutory exemption justifying the denial and release any non-exempt portions.

Thank you for your assistance. Please contact me at [email/phone] if you have any questions about this request.

Sincerely, [Name] [Contact information]

Guidance on framing

The template above is intentionally straightforward. Several considerations inform effective FOIA requests for poll tapes:

Be specific about what you want. “Results tapes” or “poll tapes” is more precise than “election records.” Specify the election, the jurisdiction, and the scope (all precincts, all voting methods). Ambiguous requests invite delay.

Know the correct law. Each state has its own public records statute with its own name, citation, and procedures. Using the correct citation signals that you know the process and are entitled to the records.

Request electronic copies when possible. Electronic copies (scanned images) are typically cheaper and faster than physical copies. They also integrate more easily into the analysis pipeline.

Be prepared for fees. Some jurisdictions charge per page or per hour of staff time. Request a cost estimate before committing to large requests.

Be prepared for delay. Response times vary enormously by jurisdiction — from same-day to months. Follow up politely if no response is received within the statutory timeframe.

Handling denials

If a request is denied, the denial should cite a specific statutory exemption. Common bases for denial include: records not yet created (if the election hasn’t occurred or tapes haven’t been processed); records exempt under a specific statute (rare for poll tapes, which are generally public records); and excessive burden (jurisdictions sometimes claim that the volume of records makes the request unreasonable).

Responses to denials depend on the state. Most states have an administrative appeal process, and some have provisions for expedited judicial review of denied public records requests. AC recommends consulting the state-specific public records law and, if necessary, seeking legal counsel.

B.13 — Version Log

This appendix records the version history of the manual.

Manual version history

v1.0. Initial release (2024). Sections 1–12 drafted and published. Stubs in state-by-state reference and voting machines reference. Substantive case study only for Monmouth NJ 2023. Several appendices as placeholders pending publication decisions.

v2.0 (current). Comprehensive rewrite. Synthesizes ~100 source documents into a single canonical reference. Major changes from v1.0:

Structural changes: Glossary moved to Appendix A. v1 Section 13 (Legal Pathways Summary) retired and absorbed into Section 6.5. Case studies expanded from placeholders to sixteen standardized entries organized by era. Appendices reorganized into three appendices (A: Glossary, B: Reference Library, C: FAQ). Six new sections added: Section 3 (Understanding Poll Tapes), Section 5 (Using the Actual Vote App), Section 9 (Doing a Comparison Analysis), Section 11 (Training and Volunteer Onboarding), Section 14 (Comparison to Other Election Integrity Approaches), Section 17 (Taxonomy of Vote Reporting Vulnerabilities). All sections renumbered accordingly; total sections expanded from 12 to 18 plus three appendices.

New material: Section 2.5 (About America Counts, expanded organizational profile), Section 2.6 (What America Counts Will and Won’t Do, governance policies), Section 4.5 (The Data Model, full formal data model with submissions taxonomy, transcribed values, comparisons, County/Precinct/Voting Method triple, comparison grain, AV Analysis lifecycle, comparison outcomes, discrepancy codes), Section 4.7 (Accessibility and Inclusion), Section 6.4 (Evidentiary Status of AV Recordings), Section 6.5 (Legal Pathways for AV Evidence, absorbing v1 Section 13), Section 6.6 (The Subpoena Question), Section 7.7 (In-App Consent Flow), Section 13.5 (AV in the Broader Election Integrity Landscape), Section 15.4 (Data Science and Measurement), B.1 (Who’s Who), B.2 (Index of AC Analyses), B.8 (AC Findings Changelog), B.11 (Election Administration Terminology Reference), B.12 (FOIA Templates).

Restructured sections: Section 13 (Philosophy) reorganized from v1’s structure into four named principles (Transparency, Non-Partisanship, Independence, Humility) plus landscape section. Section 12 (Findings and Reports) streamlined from five subsections to four. Section 1.4 (Entry Points by Reader Type) expanded from nine reader types to fifteen. Section 16 (Case Studies) expanded from placeholders and one substantive case to sixteen cases organized in four categories.

Content updates: All sections updated to reflect 2024 General Election experience and accumulated operational knowledge since v1.0.

Analyses referenced in the manual

The manual references the following Actual Vote analyses by their unique identifiers:

20240305_usa_nc_multi — North Carolina March 5, 2024 primary election. Multi-county (seven counties). Final report: NC 2024 Primary Election Report. Status: Published, canonical example.

20241105_usa_nc_multi — North Carolina November 5, 2024 general election. Multi-county. Format: rolling log on the 2024 General Election News Page. Status: Published.

20241105_usa_ga_multi — Georgia November 5, 2024 general election. Format: included in 2024 General Election News Page. Status: Published; 326 comparisons.

20241105_usa_fl_broward — Broward County, Florida, November 5, 2024 general election. Format: included in 2024 General Election News Page. Status: Published; 399 comparisons.

CAB Broward 2018 and 2020 analyses — Broward County, FL. Conducted by Citizens Audit Broward. Publication decisions pending.

Semantic versioning commitment

Future versions of the manual will use semantic versioning: v2.1, v2.2, etc. for incremental improvements (new analyses documented, corrections, clarifications); v3.0 for major restructuring. The version log will record the changes between versions. When a version is published, the previous version should be archived for reference.

B.14 — Data Schema Reference

This appendix describes the data objects that the Actual Vote system maintains, at the conceptual level. It consolidates, in one reference place, the object model that is developed functionally in Section 4 (System Architecture) and applied in Section 8 (Methodology) and Section 12 (Findings and Reports). The authoritative field-level schema will be published as part of the backend-modernization workstream described in Section 4.1. [coming soon]

The submission object

The atomic unit of input. Each submission carries: the video file(s) — the primary evidence, stored as binary blobs; a persistent integer submission ID assigned at upload (the canonical reference throughout the pipeline, citable as av.democracycounts.org/PollTape/{id}); a user account reference for chain-of-custody purposes; the recording timestamp from the device clock; GPS coordinates from device location services (where permitted), auto-resolved to city, county, state, ZIP, and country; a user-entered free-text precinct identifier (useful but unreliable — vetters and transcribers confirm the precinct from the tape header); the vetting status (Approved, Rejected, Practice, and related categories); and a free-text vetting comment. Analysis-linkage fields (which Actual Vote analysis and report a submission participated in) exist but are sparsely populated pending modernization. What is deliberately not stored matters equally: no voting history, party affiliation, demographics, or political views of the submitter, and nothing about the voters whose ballots the machine counted (see the Privacy Policy, B.5).

The vetting record

Each submission’s human review outcome: the classification (approved / rejected / practice / context / not-practical-to-transcribe, per the taxonomy in Section 4.5), the reason where applicable, privacy flags (PII detected and how it was handled), and metadata-consistency observations (GPS matching a known polling place; timestamp consistent with the election).

Transcription tuples

A transcribed value is one number read from one poll tape: a tuple of (submission ID, tape-on-submission, contest, choice, voting method, numeric value), associated with the region of the recording it was read from so that any value can be traced back to the frames that support it. A typical transcribable submission yields a few dozen to a few hundred tuples; the NC 2024 Primary’s 52 transcribable submissions produced 4,625.

Per-jurisdiction comparison logic

Each analysis carries its inferred-and-verified comparison logic as a first-class artifact: the comparison grain (typically County × Precinct × Voting Method × Contest × Choice, with documented deviations such as Georgia’s vote-center grain for Early Voting), contest/choice/precinct mapping tables reconciling tape labels with official-results labels, aggregation rules (e.g., summing multiple machines’ tapes to one precinct total), and the treatment of edge cases (write-ins, provisionals, absent official values). These are stored with the analysis (folder structure in B.10) and enumerated in its methodology notes per Section 12.6.

Comparison output

One row per comparison, with the standard fields listed in B.10 (county, precinct, voting_method, contest, choice, transcribed_value, official_value, difference, outcome, notes). The outcome label is one of exact_match, explained_match, or apparent_discrepancy (Section 4.5, Section 12.3).

Investigation history

For each initial non-match, the record of the seven-check review (Section 8.9): which checks were applied — transcription error, comparison logic error, missing poll tapes, duplicate tapes, aggregation logic error, undocumented vote reporting logic, operational context — the disposition each produced, the resulting comparison-outcome classification (match, likely attributable, apparent → escalatable → escalated → true), the assigned discrepancy code (B.10), and any correspondence with the elections office.

The field-level schema — exact field names, types, nullability, and the relational structure connecting these objects — will be published alongside the database-backed pipeline that the backend modernization workstream delivers.

B.15 — How to Re-Run Our Analysis

AC’s findings are designed to be reproducible by third parties from public materials. This appendix describes the steps. The honest formulation of what reproduction establishes is in Section 12.6; the pipeline being reproduced is Section 8.

Step 1 — Obtain the recordings. Every approved submission is public in the Actual Vote Video Archive, addressable by its submission ID (https://av.democracycounts.org/PollTape/{id}). A published analysis identifies the submissions it used; pull each one and confirm its metadata (timestamp, location) against the analysis’s scope.

Step 2 — Obtain the official-results snapshot. Official results evolve between election night, canvass, and certification. Each AC report states which official state the comparison used and preserves the archived snapshot with its capture date and provenance. Reproduce against the same snapshot; comparing against a later or earlier export is a different (also informative) exercise, but it is not a re-run.

Step 3 — Obtain the published comparison logic. For the jurisdiction and election in question, take the analysis’s documented comparison grain, mapping tables, aggregation rules, and enumerated assumptions (Section 12.6) from the published analysis package.

Step 4 — Re-transcribe or accept the transcription, then re-run the comparison. A full reproduction re-transcribes the tapes from the videos and runs the comparison logic over the result; a lighter verification accepts AC’s published transcription tuples and re-runs only the comparison step. Either way, the comparison is arithmetic: aggregate transcribed values to the grain, match against official values, and label each comparison an exact match or an initial non-match (then triaged into a match or a discrepancy).

Step 5 — Compare outputs. Check your comparison output against AC’s published output: the same comparisons, the same outcome labels, the same headline counts. Where you diverge, the divergence is itself the finding — most often it will trace to a transcription judgment or a comparison-logic assumption, both of which are enumerated precisely so they can be attacked.

The exact tooling links — the comparison scripts, packaged per-analysis inputs, and manifest documentation — will be linked here as the replication packaging of AC’s analyses is published. [coming soon]

B.16 — Security and Threat Model

AC’s verification work makes claims that depend on the integrity of AC’s own systems, and a system that audits others should publish the threat model it holds itself to. The full treatment is forthcoming with the infrastructure modernization (Section 4.1); this appendix records what that treatment must cover.

Integrity of the archive. The Actual Vote Video Archive is the evidentiary foundation. The threat model must address tampering with stored recordings or metadata, the detectability of any modification after upload, backup and redundancy, and how the archive’s integrity can be demonstrated to a skeptical third party rather than asserted.

Tampering with submissions in transit. The path from a volunteer’s device to the archive — upload authentication, transport integrity, the offline-queue-and-retry path, and the fallback intake channels (email, file sharing), which currently carry weaker integrity guarantees than in-app submission.

Insider risk. Vetters, transcribers, analysts, and administrators all touch the evidence. The model must address what a malicious or compromised insider could alter at each stage, what traces such alteration would leave, and which controls (review separation, logging, the public visibility of approved submissions) bound the damage.

Account compromise. Both volunteer accounts (could a compromised account submit fabricated recordings, and how would vetting and metadata-consistency checks catch them?) and staff/administrative accounts (which carry the insider-risk surface above).

Availability. The archive and the analysis pipeline during the windows when they matter most — election week and the before-certification window — including denial-of-service against the public archive and the operational dependency on third-party platforms (app stores, cloud services) documented in Section 4.2.

The published version will state the controls in place for each surface, the residual risks AC accepts, and the verification any outside party can perform.