2 — Conceptual Foundation

This section establishes the conceptual model that the rest of the manual assumes. By the end of it, a reader should know what vote reporting is, how it differs from vote counting, what a poll tape is, what Actual Vote does and does not do, where the vulnerabilities in the vote reporting process lie, and who America Counts is.

The six subsections build on each other in sequence. A reader already familiar with vote reporting can skip to Section 2.2. A reader who already understands the AV system and wants to assess the organization can skip to Section 2.5. Readers going cover-to-cover should read in order.

2.1 — What Vote Reporting Is

Definition

Vote reporting is the part of an election process that begins when poll tapes are printed from voting machines at a voting location, continues through the tabulation and aggregation steps performed by election officials, and ends when the county certifies the official results.

This is a narrower definition than the one most people would give if asked “how does an election get reported?” It deliberately excludes everything that happens before a poll tape comes out of a voting machine — voter registration, ballot marking, ballot scanning, the operation of the voting machine itself. It also excludes everything that happens after certification — court challenges, legislative recognition, news coverage. The definition is narrow because Actual Vote is narrow: this manual is about a tool that audits exactly this stage of the election process and no other.

Why the boundary sits where it does

The boundary between vote reporting and vote counting is the moment a voting machine prints a poll tape. Everything before that moment is vote counting. Everything after it (until certification) is vote reporting.

This boundary is not arbitrary. It marks the transition between two very different kinds of process. Vote counting happens inside voting machines — a closed, hardware-and-software process that produces numbers. Once those numbers are printed onto a poll tape, they exist as a physical artifact in the world. From that moment on, the rest of the process is human and procedural: people carry tapes from polling places to central offices, people enter numbers into spreadsheets, people aggregate and report those numbers, people sign off on totals. The work is no longer done by software inside a sealed device; it is done by people working with paper and computers.

This distinction matters because the two stages have very different vulnerabilities and very different audit requirements. Auditing vote counting requires examining voting machines, analyzing software, testing equipment, and validating that voter intent has been correctly recorded. Auditing vote reporting requires comparing what was printed on the tapes to what eventually got reported as the official total. These are different activities that need different tools.

Actual Vote is a tool for auditing vote reporting only. It does not audit vote counting. To audit vote counting you need a different tool, like Wanna Vote — a separate America Counts application that addresses the part of the election process upstream of poll tape printing. The relationship between the two is discussed in Section 2.3.

The vote reporting process step by step

A typical vote reporting process in a US jurisdiction looks like this.

Step 1: Poll tapes are printed at the voting location. When the polls close at the end of election day, the voting machines at each polling place are commanded to stop accepting ballots and to print their results. Each machine prints a strip of paper showing the vote totals it tabulated for the contests on the ballot. This strip is the poll tape. Most machines print at least two copies; many print three or more. In Texas, machines typically print five results tapes. The tape contains the voting results, the machine ID, the polling place location, and sometimes signatures from poll workers attesting that the closing procedure was performed correctly. Poll tapes are described in detail in Section 2.2.

Step 2: One copy of the tape is publicly displayed (in jurisdictions where required). In some states and some localities, one copy of the poll tape must be physically posted at the polling place after the polls close, in a location accessible to the public. The legal basis for this varies: some states have explicit statutes requiring posting, some leave it to county discretion, some don’t address it at all. In the states and localities that do require posting, the posted tape is the most accessible source of public information about the votes that were tabulated at that polling place.

Step 3: The remaining copies are delivered to the central election office. The other copies of the poll tape, along with the ballots themselves and (in most modern systems) a USB drive or memory card containing the same vote data in electronic form, are physically transported by poll workers from the polling place to the central county election office. This step is the first place where the chain of custody can be broken. The poll tapes are typically packaged in tamper-evident envelopes or boxes; the transport is typically performed by poll workers, election officials, or law enforcement; and the receiving end is typically the county elections office.

Step 4: The central office tabulates the precinct-level data. At the central county election office, the data from each polling place’s USB drives is loaded into the county’s election management software. The software aggregates the numbers from all of the county’s polling places and produces a county-level total for each contest. In many jurisdictions — Florida is an important example — the official record is the poll tape, not the USB drive, meaning that if there is a discrepancy between what the tape says and what the USB drive says, the tape takes precedence. In other jurisdictions, the USB drive is treated as the primary record. The specific rule varies by state. In practice, America Counts has documented cases where election offices rely entirely on the USB drive data and do not check it against the poll tapes — a procedural shortcut that creates exactly the kind of vulnerability AV is designed to detect.

Step 5: County totals are aggregated to the state level. Once each county has produced its own total, those totals are reported up to the state. The state aggregates them and produces a state-level total. For some elections (federal, statewide), the state-level total is the relevant outcome. For others (county or local), the county-level total is what matters.

Step 6: Results are reported publicly. At each level — county and state — the totals are published. Most jurisdictions publish unofficial results on or shortly after election night, with updates as additional ballots (provisionals, late absentees) are processed in the days following. These unofficial results are sometimes called “election night returns” or “preliminary results.”

Step 7: Certification. After a statutory waiting period (which varies by state, typically one to three weeks), the county elections board (or equivalent body) formally certifies the results. Certification is the legal act that transforms unofficial results into official results. Once certified, the results are legally final unless overturned by a court. The certification date is a critical moment in the vote reporting timeline because it marks the transition from “results that can be corrected if errors are found” to “results that can only be corrected through litigation.”

Each of these seven steps is a place where errors can occur and where fraud could in principle be committed. Each is also a place where independent observation could in principle detect such errors or fraud — but in practice, only a few of these steps are routinely subject to independent observation. The posted tape (Step 2) is observable by anyone who can get to the polling place after closing. The transport (Step 3) is observable only by the people doing the transporting. The central tabulation (Step 4) is observable by poll observers in some jurisdictions, but the observation is typically of the room, not of the data. The reported totals (Step 6) are observable by anyone but only show the final numbers, not the path that produced them.

Actual Vote works on the gap between Step 2 and Step 6. By collecting evidence of what the poll tapes said at the polling place (Step 2) and comparing it to what the official results say (Step 6), it provides an independent check on whether the steps in between were performed correctly.

The grain of vote reporting

Vote reporting in the US happens at several different grains, and understanding these grains is essential for understanding what Actual Vote can audit.

The smallest grain in most jurisdictions is the County / Precinct / Voting Method / Contest / Choice combination — the unit at which a single poll tape value is associated with a single official result value. A typical poll tape from a single voting machine in a single precinct shows, for each contest on the ballot, the number of votes cast for each choice in that contest, broken out by voting method (Election Day, Early Voting, Absentee, Provisional). Aggregated across the contests on the tape, this might be a few dozen to a few hundred values per tape.

The grain at which a jurisdiction reports its official results determines what comparisons are possible. Most US jurisdictions report at the precinct level for Election Day votes, meaning that the poll tape for a single Election Day machine in a single precinct can be matched directly against the official Election Day total for that precinct. But not all jurisdictions report at this grain for all voting methods. In some places (notably parts of Georgia), Early Voting is reported at the vote-center level rather than the precinct level, meaning that an Early Voting poll tape covering a single vote center cannot be matched against an official precinct-level Early Voting total — there is no such total to match against. In other places, Absentee votes are reported only at the county level. In yet other places (notably Wisconsin), the relevant administrative unit is the municipality rather than the county.

This means that the grain at which Actual Vote operates is not a constant. It is determined per analysis, by inspecting the structure of the official results that the relevant jurisdiction publishes for the election in question. The typical case (County × Precinct × Voting Method × Contest × Choice) is what most analyses end up using, but exceptions exist and are handled case by case. The methodology section (Section 8.7) describes the comparison logic development process in detail.

What “official results” means

The phrase officially reported results refers to the results as published by the county elections office, in whatever format and at whatever point in time. In informal contexts the manual will sometimes shorten this to official results. When the certification status matters, the manual will use uncertified official results for results published before the certification date and certified official results for results published after. Both are “official”; the distinction is whether they are legally final.

Imagine a small county called Crow County with four polling places. Each polling place has two voting machines. On election day, voters arrive, mark and feed their ballots into the machines, and the machines tabulate the votes internally. At the end of the day, the polls close and each machine prints its poll tapes. At each polling place, that’s two machines × at least two copies = at least four poll tapes per polling place, or sixteen across the county.

One copy from each machine is posted at the polling place on a wall or window. The other copies, along with the ballots and a USB drive from each machine, are sealed in a tamper-evident container and driven to the Crow County elections office.

At the elections office, the USB drives are loaded into the county’s election management software. The software adds up the numbers from each machine and produces a county-wide total. The election director publishes this total on the county’s election results page that night. Three days later, after late absentees and provisionals have been processed, the county board of elections meets, certifies the results, and the total becomes the official certified result.

If everything went correctly, the numbers on the posted tapes should add up to the published county-wide totals. If they don’t add up, something happened in between — a tape was lost, a USB drive was loaded twice, a number was entered wrong, or something more concerning occurred.

Actual Vote works by capturing the numbers on those posted tapes — through video recording — before they get taken down or weathered, and comparing them to the published totals.

2.2 — What a Poll Tape Is

A poll tape is the physical artifact at the center of the Actual Vote system. Understanding what poll tapes are, what they contain, and how they vary across jurisdictions is essential to understanding what AV can and cannot do.

Physical description

A poll tape is a strip of paper, typically thermal-printed, that comes out of a voting machine at the close of polls. It looks like a long store receipt — anywhere from a few inches to several feet long depending on how many contests were on the ballot. It is printed by the tabulating machine itself as part of its closeout procedure: when poll workers instruct the machine to finalize its count, the machine tabulates its totals and prints them.

The tape typically contains the machine ID, the precinct or vote center identifier, the election date, the names of contests on the ballot, the names of candidates or ballot measures in each contest, and the vote totals for each. Some tapes also include timestamps, poll worker signatures, a protective counter (tracking how many ballots the machine has processed in its lifetime), and various checksums or sequence numbers.

Types of poll tapes

Not all tapes are the same, and the distinction matters for AV operations.

A results tape (also called a totals tape, tally tape, or tally report) is what gets printed at the end of election day. It shows the final vote totals as tabulated by that machine. This is the tape AV is interested in.

A zero tape is printed when the machine is first turned on, before any ballots have been fed into it. It shows zeroes for every contest — confirming that the machine starts the day with no votes preloaded. Zero tapes are an important election integrity safeguard (they defend against electronic ballot stuffing), but they are not what AV compares against official results. When AV users record poll tapes at polling places, they occasionally record zero tapes by mistake, particularly when both the zero tape and the results tape are posted together. AV’s vetting process catches these.

A test tape is generated during pre-election testing of equipment. It shows the results of a known test ballot set and is used to verify that the machine is counting correctly. Test tapes are not relevant to AV operations.

The terminology varies by jurisdiction. What this manual calls a “poll tape” may be called a “printed return sheet,” a “voting machine record,” a “statement of returns,” a “tabulator totals tape,” a “votes cast form” (in California), or any number of other names. The underlying artifact is the same: a printed record of vote totals from a specific machine at a specific location for a specific election.

What a poll tape contains

A typical results tape from a precinct with a full general election ballot might include the following, in approximately this order:

The header: machine ID, election name, election date, precinct number or vote center name, and sometimes a protective counter value.

The body: for each contest on the ballot, the contest name followed by each candidate or option and the number of votes received. The body may also include undervotes and overvotes for each contest, total ballots cast, and subtotals by voting method if the machine processed multiple types (Election Day, provisional, etc.).

The footer: totals, sometimes a timestamp of when the tape was printed, sometimes a hash or checksum, and space for poll worker signatures.

The level of detail varies. Some tapes are comprehensive; others are spare. The font, spacing, and formatting also vary by machine vendor and model. This variation has practical consequences for AV: some tapes are easy to read in a video recording, and some are not.

Posting requirements

Whether and how poll tapes are publicly posted after polls close varies by state and sometimes by county. In states with posting requirements, one copy of the results tape must be displayed at the polling place in a location accessible to the public — typically taped to a window or door. The legal details are covered in Section 6 and the state-by-state reference in B.3.

In practice, even in jurisdictions with clear posting requirements, compliance is uneven. America Counts has documented cases of tapes posted behind locked doors at schools and churches after hours, tapes posted behind glass with glare that defeats recording, tapes folded or taped together so that vote totals are obscured, tapes posted upside-down, and tapes simply not posted at all despite a legal requirement to do so. These posting failures are documented as operational issues in Section 10.11 and are themselves a finding — even when no reporting discrepancy is detected, the failure of election officials to comply with posting requirements is worth documenting.

Quality issues

Poll tape quality is a recurring challenge for AV operations. Several factors affect whether a recording of a poll tape can be used for comparison analysis.

Ink density. Thermal printers produce fainter output as they age or as paper quality varies. Faint printing makes video transcription difficult or impossible.

Font legibility. The font used by some voting machines makes certain digits hard to distinguish from each other. The most notorious example is the Proxima Nova font family, where the digits 6 and 8 are nearly identical at the resolution of a typical video recording. America Counts recommends that voting equipment manufacturers and election administrators use high-legibility fonts where the digits 1/I, 6/8, and 0/O are clearly distinguishable. (For its own internal working spreadsheets, AC uses IBM Plex at 12pt for this reason.)

Tape length and condition. Long tapes can be unwieldy to post and difficult to record in a single video. Tapes that have been crinkled, torn, or exposed to weather may be partially unreadable. Tapes that have been cut by election officials — a practice AC has documented — may be missing portions of their content.

Chain of custody from printing through posting. The time between when a tape is printed and when it is posted is a window during which the tape could in principle be altered or replaced. AV’s video recordings establish a point-in-time record of what the tape showed at the moment of recording. This is part of the evidentiary value of the AV system: the GPS-stamped, timestamped video recording creates an independent record that exists outside the election administration’s chain of custody.

2.3 — What Actual Vote Does and Doesn’t Do

This section is critically important. America Counts is committed to honesty about what its tool can and cannot accomplish, and the manual is committed to setting expectations precisely. A reader who finishes this section should have a clear sense of what Actual Vote will and will not detect, what other tools exist for the things it does not do, and what the residual risks are even when an Actual Vote analysis returns a clean result.

What Actual Vote does

Actual Vote compares poll tape values against officially reported results to detect discrepancies in the vote reporting layer.

Concretely: AV users video-record poll tapes. America Counts transcribes the vote totals from those recordings into structured data. AC then obtains the officially reported results for the same election from the relevant elections office. AC compares the two datasets, value by value, at the finest grain available. Where the values match, vote reporting appears accurate for those values. Where they do not match, there is an initial non-match that review then triages.

The result of this process is an Actual Vote analysis — a report documenting which values matched, which did not, and what investigation of any discrepancies revealed. The system architecture is described in Section 4 and the full methodology in Section 8.

What Actual Vote does not do

AV does not verify vote counting. Actual Vote does not examine whether the voting machine correctly counted the ballots fed into it. If a voting machine tabulated incorrectly — due to a hardware fault, a software bug, or a deliberate attack — those incorrect totals will be printed on the poll tape just as faithfully as correct totals would be. Actual Vote will then verify that the incorrect totals were faithfully transcribed into the official results — a successful comparison, a match — and the underlying error in vote counting will go undetected.

This is the matched-poll-tape caveat, and it is the most important single limitation of Actual Vote: if the poll tapes themselves already reflect error or fraud, but the vote reporting from those tapes is done correctly, an Actual Vote analysis will say “vote reporting appears accurate.” The manual restates this caveat in several places because it is critical to understanding what AC’s findings mean.

AV does not audit voter eligibility or voter access. Actual Vote has nothing to say about whether the right people were allowed to vote, whether eligible voters were turned away, whether voter rolls were maintained correctly, or whether any kind of voter suppression occurred. An election can be reported with perfect accuracy from a vote-reporting standpoint while still failing to reflect the will of the eligible voting population. AV will confirm the reporting was accurate; it cannot confirm the election was fair.

AV does not audit voting machine integrity. It does not check whether tabulators correctly counted ballots, whether optical scanners correctly read ballot marks, whether ballot-marking devices correctly recorded selections, or whether malicious code was running on any election equipment. This is a particularly important limitation because the most prominent public concerns about election integrity in recent years have been about voting machines specifically.

AV does not verify individual ballots or individual votes. AV works at the level of aggregated totals on poll tapes. It does not see individual ballots. It cannot say whether any specific voter’s ballot was correctly recorded. It operates entirely above the level at which individual voter data exists — a feature that is essential to its independence, since auditing individual ballots would require either being inside the chain-of-custody system (compromising independence) or violating ballot secrecy (compromising voter privacy).

AV does not make legal determinations about causes. When AV identifies a discrepancy between a poll tape value and an official result, it reports the discrepancy. It does not characterize the discrepancy as “fraud” or assign intent. The discrepancy could be a clerical error, a software bug, a procedural failure, or something more concerning. AV identifies the symptom — the mismatch — and leaves the diagnosis to investigation.

The matched conspiracy caveat

A determined attacker who controlled both the printing of poll tapes and the data delivered to the central tabulation office could in principle alter both in identical ways, producing a poll tape that matches the (also-altered) official record. Such an attack would defeat Actual Vote because the comparison would still succeed.

This is a real limitation, and the trustworthiness discussion in Section 13 addresses it directly. The argument there is roughly: a matched-conspiracy attack is much harder than the attacks Actual Vote does detect, because it requires coordinated control of multiple independent steps in the process. The threat actors who could plausibly conduct such an attack have easier options available. So the matched-conspiracy caveat bounds AV’s scope but does not undermine its value.

The manual takes this caveat seriously and will not pretend it doesn’t exist. But the caveat does not mean Actual Vote is useless. AV is designed to detect a specific class of attack (and a specific class of unintentional error), and it is effective against that class. The fact that other classes exist is a reason to use Actual Vote alongside other tools, not a reason to dismiss it.

Relationship to Wanna Vote

Wanna Vote is a separate America Counts application, currently in development, that addresses the vote counting side of the election process — the part upstream of poll tape printing that AV does not cover.

Where AV asks “were the numbers on the poll tapes faithfully reported as official results?”, Wanna Vote asks “did the voting machines correctly count the ballots that voters cast?” It does this through citizen canvassing surveys: volunteers survey voters about how they voted, building an independent ground-truth dataset that can be compared against official results. This is methodologically distinct from what AV does. AV works with physical artifacts (poll tapes) and official data (published results). Wanna Vote works with survey data (reported voter intent) and official data.

Together, AV and Wanna Vote would cover both halves of the verification problem: reporting accuracy (AV) and counting accuracy (Wanna Vote). But they are separate tools with separate methodologies, and this manual is about AV. Wanna Vote is mentioned where the distinction clarifies AV’s scope and is not discussed further.

Operational limitations

Beyond the conceptual limitations above, Actual Vote has practical operational limits.

Tapes that aren’t posted cannot be recorded via the primary access method. In jurisdictions where poll tapes are not publicly displayed after polls close, the most common recording method (recording at polling places on election night) is not available. Other access methods — poll observer recording, in-office viewing, FOIA requests — are available but require more effort and produce results on a different timeline. Access methods are covered in detail in Section 7.

Tapes that are posted incorrectly may not be usable. America Counts has documented numerous failure modes in posted tapes: tapes behind glass with glare, tapes folded so the totals are not visible, tapes posted upside-down or with key information obscured, tapes torn or weathered, tapes that have been cut or taped together. Each of these reduces the usefulness of any recording.

Recording quality varies. Even when a tape is properly posted, the recording may be too blurry, too dark, too short, or too shaky to be usable. America Counts vets every submission and rejects unusable ones. Each rejection means a piece of potential evidence is lost.

Some jurisdictions report at grains that don’t match poll tape grains. As discussed in Section 2.1, some voting methods are reported at higher grains than the poll tapes that produce them. Where the grains don’t align, no comparison is possible for those values.

Some jurisdictions actively obscure their official results. At least one large North Carolina county publishes its official results in a PDF format that uses non-standard character encoding, hidden positioning offsets, and image-substituted text — properties that effectively defeat machine-readable extraction. AC can still obtain the results through manual transcription, but the effort is substantially greater. Such obstacles slow an analysis without preventing it.

The pipeline takes time. From recording to publication of an Actual Vote analysis report can be days to weeks, depending on submission volume, comparison logic complexity, SOE responsiveness, and available transcription help. America Counts cannot guarantee that any specific submission will be analyzed before the certification date of the relevant election.

The User Agreement: no guarantees

America Counts is explicit in its User Agreement about the limits of what it can promise. The agreement states that AC cannot guarantee that any specific submission will be transcribed and analyzed before the certification date, that an apparent discrepancy for which AC asks the SOE will receive a satisfactory response, that a court challenge based on AV evidence will be mounted or will succeed, or that the software will work on every device under every condition. The 2024 General Election experience with the Google Play store rejection — which delayed the latest Android version until after election day — is a documented case of this last limit being reached in practice.

These are not weaknesses of the system; they are honest acknowledgments of what a small nonprofit can and cannot do. The User Agreement is reproduced in full as Appendix B.4.

2.4 — Vulnerability Points in Vote Reporting

Vote reporting in the US is vulnerable because it is a long, multi-step, human-and-software process performed by people working under time pressure with limited oversight. Errors are common, fraud is possible, and the existing audit mechanisms cover only some of the failure modes. This section describes the threat landscape — what kinds of error and fraud Actual Vote is designed to surface, and why those threats are real.

This is a brief overview. For the full treatment — eleven sourced categories of vote-reporting failure, each with documented incidents, structurally plausible variants, and a per-category account of what Actual Vote does and does not catch — see the Taxonomy of Vote Reporting Vulnerabilities. That detailed taxonomy lives alongside the Case Studies rather than here in the manual, because it is built directly from them: the case studies supply the evidence, and the taxonomy generalizes it.

Four categories of vulnerability

America Counts organizes vote reporting vulnerabilities into four categories, based on where in the process the error or manipulation occurs.

Data entry errors. When poll tapes or USB drives arrive at the central election office, the numbers must be loaded into the election management system. This loading can be automated (from USB drives) or manual (from paper tapes). In both cases, errors can occur: a USB drive loaded twice, a number transposed during manual entry, a tape misread. These are the most mundane kind of error and the most common.

Aggregation errors. Once precinct-level data is in the election management system, the software aggregates it to produce county-level and state-level totals. Aggregation errors occur when the software adds the numbers incorrectly — typically because of a bug, a configuration error, or an incorrect mapping between precincts and reporting units. The Monmouth County, New Jersey case is a clean example: a software reinstallation bypassed the safeguard against duplicate USB drive loading, causing votes from one location to be counted twice. The error was enough to change the outcome of a close race.

Export and formatting errors. Even when internal data is correct, the act of publishing results can introduce errors. A PDF generated with incorrect formatting, a CSV file with truncated fields, a web page that displays totals from an earlier data snapshot — all of these can produce a gap between what the election office knows internally and what the public sees. The Wake County, North Carolina 2022 case is an extreme example: official results PDFs contained deliberate obfuscation through non-standard character encoding and hidden positioning, making machine-readable extraction effectively impossible.

Reporting errors. The broadest category: any case where the number that appears in the officially published results differs from the number that was on the poll tape. This includes errors caused by any of the first three categories, and also includes cases where the cause is unclear or where the discrepancy cannot be traced to a specific step.

Unintentional versus intentional

America Counts investigates the discrepancy, not the motive. When AV identifies a gap between a poll tape value and an official result, it reports the gap. It does not — and usually cannot — determine whether the gap resulted from human error, software malfunction, or deliberate manipulation.

Unintentional errors are by far the more common finding. They happen because the process is complicated, the people are tired (poll workers routinely work 14-hour days), and the equipment occasionally misbehaves. Unintentional errors are not headline-grabbing, but they can change the outcomes of close races and they cumulatively erode the accuracy of the public record.

Intentional fraud in vote reporting is rarer but not unheard of. Fraud is harder to detect because it is designed to evade detection. The categories include tape interception or destruction during transport, malicious code in election management software, physical access to election office computers, and insider attacks by election officials with the access necessary to alter results. The manual is not in the business of accusing anyone of fraud. The point of listing these vectors is to acknowledge that they exist as theoretical possibilities and that any audit system worth its name should be designed to detect them when they occur.

Actual Vote is agnostic about intent. It detects discrepancies. Whether a discrepancy resulted from a tired poll worker or a deliberate manipulation, the discrepancy is equally worth identifying and investigating.

Real-world examples

The threat landscape is not theoretical. Vote reporting errors have been documented in real elections with real consequences. Three brief examples illustrate the range; detailed treatment appears in the Case Studies section.

Shelby County, Tennessee 2015. A citizen named Bennie Smith photographed a poll tape at the Unity Christian Church precinct in Memphis. The tape showed 546 votes, but the official results reported only 330 — roughly 40% of votes from this predominantly Black precinct had disappeared during the reporting process. Investigation revealed that the GEMS tabulator software being used had known bugs dating back to at least 2008 in Ohio that had never been patched. The missing votes were eventually recovered, but the corrective analysis was suppressed. Had AV existed and been deployed, the discrepancy would have been detected and permanently documented.

Monmouth County, New Jersey 2023. An ES&S voting system double-counted votes from one polling location due to a software reinstallation that bypassed duplicate detection. The error flipped a candidate from third to fourth place by a single vote. The error was discovered, acknowledged by the county and the vendor, and corrected. This case demonstrates that modern voting systems with current technology still produce serious reporting errors, and that independent verification is the backstop.

Wake County, North Carolina 2022. Official results PDFs contained deliberate obfuscation: nuisance characters, encoding tricks, hidden columns. Whether this reflected an intent to hide data or a misguided implementation of “security through obscurity” is unclear. Either way, the effect was to make independent verification of the published results substantially more difficult. AV bypasses this kind of obstruction by working from the primary source — the poll tape — rather than from the published results format.

The spectrum from benign to concerning

These examples illustrate a spectrum. At one end, a tired poll worker transposes a digit while entering numbers at 2 AM — a benign error that produces a small discrepancy likely caught during canvassing. At the other end, systematic manipulation of tabulator software could produce large discrepancies that would go undetected without independent verification. AV detects the symptom (the discrepancy between poll tape and official result) regardless of where on this spectrum the cause falls.

America Counts does not position itself as primarily concerned with fraud. The organization’s public position is that errors in vote reporting — whether intentional or not — undermine confidence in election outcomes, and that independent verification benefits everyone regardless of which theory of election risk a reader subscribes to. A reader who believes most errors are innocent mistakes will still value AV as a quality-control check. A reader who worries about deliberate manipulation will value AV as a detection tool. Both readings are valid.

Why traditional audit mechanisms leave gaps

The existing audit mechanisms have value, but each has structural limitations.

Risk-Limiting Audits (RLAs) are well-respected statistical procedures that examine a sample of ballots to verify that the reported winner is correct. But they require election authority cooperation, which is not always forthcoming; they focus on verifying the winner rather than the totals; and they address vote counting, not vote reporting. AV and RLAs are complementary — they cover different parts of the process. The relationship between AV and other tools in the election integrity landscape is discussed in Section 13.5.

Recounts are after-the-fact reviews typically run by the same officials who ran the original count. They cannot independently verify the original work and typically do not address reporting-layer errors.

Election observation by party representatives provides some oversight at polling places but usually does not extend to the central tabulation step where many reporting errors occur.

Court challenges can correct errors but require evidence — and the evidence has to come from somewhere. Without a tool like AV, the evidence available to a candidate considering a challenge typically consists of secondhand reports, anonymous tips, and statistical anomalies. AV provides the missing link: primary-source, video-based, independently collected evidence of what the poll tapes actually said.

There is also a simpler reason the reporting layer has gone unchecked: until recently, checking it at any scale was infeasible. A single vote-center tape can run the entire county ballot — well over a hundred contests — and verifying one tape against official results by hand means transcribing thousands of thermal-printed numbers without an error. That does not scale, which is why independent confirmation of vote reporting has historically not happened at any meaningful volume. The premise of Actual Vote is that this is now a throughput problem, not just an accuracy problem, and that throughput problems can be engineered away. A volunteer points a phone at the posted tape; the system reads every number off it and checks it, contest by contest, against the certified results; and only discrepancies surface for a person to look at. A phone video goes in and an independent, citable audit comes out — cheaply, quickly, and repeatably enough to do at scale. The engineering that makes this real is described in Section 4.8; what matters here is the shift it represents: a check that was once impractical for anyone to perform is now practical to perform everywhere.

See our Case Studies section for documented examples, and its taxonomy of vote-reporting vulnerabilities.

2.5 — About America Counts

America Counts is the organization behind Actual Vote. It is small, it is nonprofit, it is non-partisan, and it has been at this work for over a decade. This section introduces the organization for readers who are encountering it for the first time and provides enough depth for readers — particularly lawyers, donors, and potential partners — to assess the organization’s credibility and maturity.

Organizational identity

Democracy Counts, Inc. is a 501(c)(3) registered non-profit, non-partisan corporation. America Counts is the public-facing name for the organization’s election audit initiative. Both names appear in AC publications: “Democracy Counts” in legal contexts (copyright notices, donation acknowledgments, IRS filings) and “America Counts” in public-facing contexts (the website, the app, this manual). Actual Vote is a project of America Counts.

The 501(c)(3) status is significant because it imposes legal constraints. As a 501(c)(3), America Counts cannot endorse candidates, cannot engage in partisan political activity, and is restricted in the kinds of advocacy it can undertake. These constraints happen to align with America Counts’ mission, which is non-partisan by design. But the legal constraints provide an additional layer of accountability: America Counts is not merely choosing to be non-partisan — it is legally required to be.

Origin story

Dan Wolf’s path to founding Democracy Counts was indirect but coherent. As a student at Harvard Law School, he spent the summer of 1984 in Nicaragua researching their electoral system, producing what was the world’s first country-specific manual for election monitors. He continued research in Nicaragua and El Salvador through the 1980s, receiving a Fulbright Fellowship for his dissertation research in political science at the University of California, San Diego. In 2012, he served as senior adviser to the International Election Observation Mission that monitored the Taiwan elections.

Between the academic work and Democracy Counts, Wolf pursued other ventures: founding the TransBorder Institute at the University of San Diego, creating Terra Segura International (an NGO that developed landmine-clearing technology), and building two technology companies — one that further developed the landmine technology and another that developed unmanned ground vehicles for use against improvised explosive devices.

The pivot to election technology came from an unlikely source: SEC audits. As CEO of a public company, Wolf became intimately familiar with the rigor of independent financial auditing required by the Securities and Exchange Act. The experience crystallized a parallel: if the public interest in transparent finance justifies the rigor of SEC oversight, then the public interest in transparent elections justifies equal rigor. Public companies cannot audit themselves. Why should elections be any different?

Wolf recognized that mobile technology could enable something previously impractical: citizen-powered, independent, scalable auditing of election results. The idea was not to replace official processes but to supplement them — to create an independent check that operates outside the election administration’s chain of custody. In 2015, he founded Democracy Counts to build the tools.

Key personnel

Dan Wolf is the founder and CEO of Democracy Counts. Wolf is the architect of the organizational strategy, the author of the foundational 2018 Verify the Vote white paper, and the public face of America Counts in many contexts — interviews, advocacy meetings, donor relations, and strategic partnerships. His background in law, political science, international election observation, and technology entrepreneurship shapes the organization’s approach: methodologically rigorous, legally aware, technologically enabled, and built for eventual scale.

Jason Flatley is the CTO and runs analysis operations. Flatley is the person who actually executes the Comparison Analysis pipeline — building per-jurisdiction comparison logic, vetting submissions, doing transcription work, investigating apparent discrepancies, writing reports, and handling the data science that underlies targeting and impact measurement. Flatley developed the Actual Vote methodology as it exists today, oversees app development, and serves as the operational contact for analysis questions and field coordination. The default contact for technical inquiries is j.flatley@democracycounts.org.

Beyond the core team, America Counts works with occasional paid contractors (for specific projects and field deployments), partner organizations (most notably Scrutineers, which connects AC with volunteers), and a distributed network of volunteers who contribute in many ways — from recording poll tapes to helping with transcription to assisting with research and outreach. The organization has consistently described itself as “a very small volunteer organization with a very tight budget.” This is factual accuracy about constraints, not a complaint.

Organizational history

The trajectory of America Counts can be traced through a series of milestones.

2015–2017: Founding and early development. Wolf founds Democracy Counts and begins developing the Actual Vote concept and the supporting mobile application.

2018: The Verify the Vote white paper. Published as a special report for Democracy Counts, Verify the Vote lays out the philosophical case for citizen-driven independent auditing of vote reporting. It articulates the SEC analogy, the theory of change, and the case for why existing audit mechanisms are insufficient.

2018–2020: Broward County, Florida field operations. America Counts conducts its first major field operations in Broward County, in partnership with the local group Citizens Audit Broward (a bipartisan, professionally organized audit committee). The work documents extensive procedural violations — poll tapes posted behind windows, obscured, taped together, missing — while finding no reporting errors in the analyzed precincts. This early work establishes the operational pattern: the absence of reporting errors is itself a finding, and procedural problems are worth documenting regardless.

2021: Georgia runoff deployment. AC deploys to cover the January 2021 Georgia Senate runoff elections, expanding geographic reach beyond Florida.

2022: North Carolina operations and Wake County discovery. AC begins operations in North Carolina. The Wake County analysis reveals that official results PDFs contain deliberate obfuscation, an important finding about transparency practices that informs subsequent work.

2024: NC Primary — the headline result. The North Carolina March 5, 2024 Primary analysis covers seven counties and 321,502 individual votes. The finding: zero true discrepancies. This is the most complete AV analysis to date and the manual’s canonical example of the full pipeline operating at scale.

2024: Multi-state General Election deployment. AC deploys across North Carolina, Georgia, Florida, and additional states for the November 2024 General Election. This is the largest operational deployment to date and produces the most extensive operational lessons (documented in Section 10.11), including the Google Play store rejection that delayed the Android app launch until after election day.

The SEC analogy

A recurring theme in America Counts’ philosophy is the parallel between financial auditing and election auditing.

Public companies in the United States are required by law to submit to independent financial audits — not because regulators assume companies are dishonest, but because the public interest in transparent financial markets justifies the cost and rigor of independent verification. The same logic applies to elections. Elections determine who governs. The public interest in knowing that election results are accurate is at least as compelling as the public interest in knowing that corporate financial statements are accurate. Yet elections have no equivalent to the independent audit requirement. Election administrators, in effect, audit themselves — or are not audited at all.

America Counts aims to fill that gap. The organization does not claim to be the SEC of elections; it does not have regulatory authority and does not seek it. But it provides the independent verification function that the SEC analogy suggests should exist: an outside entity, operating independently of the system being verified, checking the numbers against primary sources.

Structure and funding

America Counts is funded primarily through individual donations to Democracy Counts, Inc. As a 501(c)(3), donations are tax-deductible to the extent permitted by law. The organization does not accept donations conditional on partisan outcomes and does not accept donations from political parties or political action committees.

The total budget is small. The team has long been accustomed to operating with constraints that would be unfamiliar to better-funded organizations. Much of the work is done by volunteers; paid contractor and paid worker deployments are reserved for specific tasks where dedicated time is essential (the 2024 Georgia paid worker deployment being the most substantial example to date).

Microsoft has donated cloud services and technical support; the Actual Vote app and website backend run on Microsoft Azure. These in-kind donations have been significant in allowing the organization to maintain its technical infrastructure at a fraction of commercial cost.

Sustainability and growth

America Counts has explored several models for long-term sustainability beyond the current philanthropic base.

Data partnerships. The poll tape data AC collects has potential commercial value. Firms like Edison Research, which provide election-night projections to media organizations, rely on exit polls and sample data to make early calls. AC’s poll tape data — which arrives faster than official results and is more granular than exit poll projections — could supplement those projections. This remains conceptual but represents a plausible revenue path that does not compromise independence.

Technology licensing. The methodology and infrastructure AC has developed could be licensed to other election integrity organizations, both domestic and international. A for-profit subsidiary structure has been explored as a vehicle for this.

Grant funding at scale. AC has pursued foundation support for major deployments. The ambition is substantial: scaling Actual Vote to cover a significant portion of America’s approximately 113,000 polling places would require resources well beyond the current budget, but the impact would be proportionally transformative.

Volunteer-to-paid-worker pipeline. The 2024 Georgia deployment demonstrated a model where experienced volunteers transition to paid field workers for high-priority deployments, combining institutional knowledge with the reliability that compensation brings.

Future vision

America Counts’ longer-term vision extends in several directions.

Professionalization of election auditing. Wolf has proposed the creation of an American Federation of Election Auditors — a body that would develop standards and best practices for independent election audits, train and certify election auditors as a recognized profession, and provide a forum for consensus positions on election integrity issues. This idea has been met with interest from other organizations in the election security community but remains in the conceptual stage.

Comprehensive verification. AV covers vote reporting; Wanna Vote (in development) covers vote counting. Together with additional tools addressing voter registration accuracy and ballot chain-of-custody verification, the full suite would cover most of the opportunity space for election error and fraud. Democracy Counts’ growth strategy is to focus on one tool at a time, achieve acceptance, and build on that for the next.

International expansion. Nothing about the AV methodology is US-specific. The principle — independent citizens recording publicly accessible election results and comparing them against official reports — applies wherever elections produce paper records and official reporting. Unsolicited expressions of interest have come from Mexico, Nicaragua, Gabon, Kenya, and Pakistan. International deployment would require adapting to local legal frameworks and election procedures, but the core methodology transfers.

How to engage with America Counts

To use the Actual Vote app: download it from the Apple App Store or Google Play Store and follow the in-app onboarding. The Get Started section of americacounts.us provides a walkthrough.

To volunteer: visit americacounts.us or contact Scrutineers (scrutineers.org), AC’s partner organization for volunteer connection.

To donate: visit democracycounts.org for current donation options.

To contact AC about a technical or analysis question: email j.flatley@democracycounts.org.

To partner on an analysis in your jurisdiction: contact Jason Flatley directly. AC welcomes partnerships with local civic organizations, candidates, and election integrity groups across the political spectrum, subject to the non-partisan operating principles.

2.6 — What America Counts Will and Won’t Do

The previous section described what the Actual Vote system does and does not do — its technical scope. This section describes what the organization will and will not do — its governance policies. These are distinct. The system has technical limitations (Section 2.3); the organization has principled commitments.

Non-partisan positioning

America Counts will not take partisan donations, endorse candidates, or cooperate with campaigns that want to use AV findings as a partisan weapon. This commitment is structural, not merely aspirational: the 501(c)(3) legal framework prohibits partisan political activity, and the organization has built its credibility on consistent non-partisan behavior.

In practice, this means AC reports findings regardless of which party or candidate they appear to favor. When analyses have found no discrepancies — as has been the case for the great majority of analyses to date — AC reports that finding straightforwardly, even though clean results are less interesting to media and partisans than the alternative would be. When analyses do find discrepancies, AC reports them with equal directness, regardless of who benefits. This consistency is what allows the organization to maintain credibility across the political spectrum.

The non-partisan commitment is tested most acutely by people who approach AV with an existing theory about which party is cheating. Strong partisans on both sides tend to arrive expecting AV to confirm their priors. AC’s response is the same to both: AV follows the evidence. If the evidence confirms your suspicions, you’ll know. If it doesn’t, you’ll know that too. The data is public; anyone can check.

Language discipline

America Counts will not characterize discrepancies as “fraud.” When AV identifies a gap between a poll tape value and an official result, AC will call it a “discrepancy,” an “error,” or an “unexplained difference” — saying only what the tapes show, and leaving characterization of intent to investigation and, if warranted, to legal proceedings. The full statement of this conservative-voice principle lives in the Philosophy chapter (Section 13.4, the Humility Principle).

Cooperation and independence

America Counts will work with election administrators who engage in good faith. The Henderson County, North Carolina model (described in Section 8.11) exemplifies the preferred approach: AC contacts the SOE with findings, the SOE provides clarifying information, discrepancies are resolved collaboratively. This is the best outcome for everyone — the SOE learns about potential issues, AC gets the context needed to interpret its findings, and the public record is accurate.

At the same time, AC will not compromise its independence to maintain a cooperative relationship. If an SOE refuses to engage, stonewalls, or provides information that contradicts the evidence, AC will document that too. Independence means being willing to publish findings that make election administrators uncomfortable when the evidence warrants it.

America Counts will share its methodology and findings transparently. It will cooperate with legitimate researchers and journalists. It will provide data to anyone who requests it for legitimate purposes, subject to the privacy protections described in the Privacy Policy.

What AC shares publicly: all approved recordings (in the Video Archive), all published analysis reports, the general methodology, and aggregate findings.

What AC shares with researchers and journalists upon request: detailed analysis workbooks, transcription data, comparison logic documentation, and supporting materials for published findings.

What AC keeps private: raw submissions prior to vetting, user account information, internal communications, draft reports prior to publication, and any personally identifiable information captured incidentally during recording. The Privacy Policy (reproduced as Appendix B.5) describes data handling in full detail.

The corrections policy

America Counts will acknowledge and correct its own errors. If AC discovers a transcription error, a comparison logic error, or an analytical mistake in a published report, it will update the report with a correction, document the error and its resolution, and note the correction in the Findings Changelog (B.8).

This corrections policy is a deliberate commitment to intellectual honesty. In the election integrity space, where many organizations have published claims based on errors they later refused to acknowledge, AC believes that visible self-correction builds more credibility than an appearance of infallibility. Being wrong is inevitable in complex analytical work. What matters is whether you fix it.

Monetization boundaries

America Counts will not pursue monetization models that compromise its independence or its non-partisan positioning. Any future commercial activities (data licensing, technology licensing, for-profit subsidiary) will be structured to preserve the independence and credibility of the core audit work. Revenue-generating activities will supplement, not replace, the philanthropic and volunteer base that gives the organization its independence from commercial pressures.