Appendix A — Glossary
This glossary collects the formal definitions used throughout the manual into a single alphabetized reference. Each entry provides the term, its definition, and a link to where the term is substantively discussed. Readers encountering an unfamiliar term anywhere in the manual can look it up here.
A
Access method. One of five canonical paths by which an Actual Vote user obtains poll tape recordings. The five methods are: outside polling places on election night (Method 1), poll observer or poll worker recording during tabulation (Method 2), appointment with the local elections office (Method 3), Freedom of Information Act or public records request (Method 4), and special cases (Method 5). Each method derives from one of three legal cases described in the Legal Basis chapter. Learn more here.
Actual Vote. America Counts’s free iOS and Android app for video-recording poll tapes and uploading them to the Actual Vote Video Archive for analysis. Often also used as shorthand for the broader Actual Vote system — the app together with the archive, the guidance, the analysis pipeline, and the publishing layer. The manual generally specifies which meaning is intended when ambiguity could arise. Learn more here.
Actual Vote analysis. One complete unit of analysis work, scoped as one specific election in one specific state, with all the submissions, transcriptions, comparisons, investigations, and findings that go into producing one report. Each Actual Vote analysis has a unique identifier of the form [date]_[country]_[state]_[scope]. Learn more here.
Actual Vote recording project. A planned effort — by one person or an organized group — to video-record poll tapes in a particular area for a particular election, using the Actual Vote app. A recording project is the field work that produces submissions; those submissions may then feed one or more Actual Vote analyses. Distinct from an Actual Vote analysis (the downstream comparison of one election in one jurisdiction) and from a Comparison Analysis (the analytical pipeline). Replaces informal uses of “Actual Vote project.” Learn more here.
Actual Vote system. The whole ecosystem America Counts builds around poll-tape auditing — more than just the app. It comprises AC’s knowledge base about poll tapes and vote-report auditing; its up-to-date, area-specific guidance on how to use the app effectively; its help planning individual and group Actual Vote recording projects in states across the country; the Actual Vote app itself; and the high-throughput, AI-assisted transcription, comparison, and reporting software that turns the hard work of recording poll tapes into greater election transparency. Learn more here.
Actual Vote Video Archive. The public-facing web application at https://av.democracycounts.org/ that hosts all approved Actual Vote submissions. Each submission has a permanent integer ID and a canonical URL of the form https://av.democracycounts.org/PollTape/{id}. Learn more here.
Aggregation rule. A rule, specific to a jurisdiction × election, that specifies how multiple transcribed values are combined before being compared against a single official value. The most common case: summing values from multiple machine tapes for the same precinct × voting method before comparing against the precinct total. Learn more here.
America Counts (AC). The non-profit non-partisan election audit initiative that operates the Actual Vote system. America Counts is an initiative of Democracy Counts, Inc., a 501(c)(3) registered non-profit corporation. Founder and CEO: Dan Wolf. CTO: Jason Flatley. Learn more here.
Analysis ID. The unique identifier assigned to one Actual Vote analysis, of the form [date]_[country]_[state] with an optional scope suffix — e.g. 20240305_usa_nc for the 2024 North Carolina Primary analysis. Used to identify the analysis consistently across the archive, the transcription datasets, and the published report. Learn more here.
Analysis status. Whether an Actual Vote analysis is open (still accepting submissions, transcribing, comparing, or investigating) or closed (all comparisons and investigations complete; findings final as published). A closed analysis is modified only by a formal revision, recorded in the report’s revision history.
Analysis Window. The time period during which an Actual Vote analysis is conducted. Two windows are distinguished: the before-certification window (before the relevant jurisdiction has certified the election results) and the after-certification window (after certification). Findings produced before certification have operational urgency; findings produced after certification serve the historical record. Learn more here.
Apparent discrepancy. A discrepancy for which AC’s review produced no benign hypothesis about the vote-reporting logic that would resolve it. Because the unexplained cause could be innocent error or fraud, obtaining an answer is urgent. An apparent discrepancy is the only gateway to a true discrepancy — no confirmed reporting error is reached without first being unexplained. It can be of any magnitude and may be contest-flipping or not. Exits: an AC investigation or a heavier ask to the SOE yields a satisfactory resolution → match; the SOE acknowledges the error → true discrepancy; no satisfactory resolution → escalatable. The “apparent” qualifier avoids pre-judging what investigation will find. Learn more here.
Approved submission. A submission that has passed America Counts’s vetting process and is publicly visible in the Actual Vote Video Archive. See also: submission. Learn more here.
Ask to the SOE (lighter and heavier). AC’s cooperative inquiry to the elections office when a discrepancy needs outside information. Its weight is proportional to how unexplained the discrepancy is: a likely attributable discrepancy warrants a lighter ask (we hold a benign hypothesis and just want confirmation); an apparent discrepancy warrants a heavier ask (we have no explanation, so the inquiry is more formal). Asking the SOE is cooperative and is never called escalation — that word is reserved for handing a matter to a court-capable party. Learn more here.
Audit committee. A category of Actual Vote user: a local civic group that operates the full Comparison Analysis pipeline themselves rather than just contributing recordings to AC’s central pipeline. The Citizens Audit Broward (CAB) effort is the canonical example. Learn more here.
Average votes per transcribed value (AVTV). Total votes audited divided by the number of transcribed values in an analysis. Imagine an Actual Vote analysis covering a single poll tape with a single race: transcribing 10 votes for Candidate A and 20 for Candidate B gives 30 total votes across 2 transcribed values — an AVTV of 30/2 = 15. If the tape instead read 1,000 and 2,000, the AVTV would be 3,000/2 = 1,500. AVTV measures how much vote-report-auditing potential each transcribed value yields on average. This matters because transcription takes effort: higher-AVTV analyses produce more total votes audited for the same amount of transcription work. 69.5 in the NC 2024 Primary Report.
Average votes per transcribable submission (AVTS). Total votes audited divided by the number of transcribable submissions in an analysis. AVTS measures how much vote-report-auditing potential each transcribable submission yields on average. This matters because recording poll tapes with the Actual Vote app takes effort: higher-AVTS analyses produce more total votes audited for the same amount of recording work.
B
Ballot. The physical or electronic document on which a voter records their choices. AV does not interact with ballots directly; AV interacts with poll tapes, which are the machine-produced summaries of ballots counted. The distinction between the ballot (the voter’s expression) and the poll tape (the machine’s count) is central to understanding AV’s scope.
Ballot image audit. A post-election review that examines digital images of scanned ballots rather than the physical ballots themselves. Some scanners can be configured to save an image of each ballot as it is scanned. Ballot image audits check counting-layer accuracy (whether the scanner interpreted marks correctly) and are complementary to AV, which checks reporting-layer accuracy. Learn more here.
Ballot jam re-scan. An operational event in which a paper ballot jams during tabulation, is cleared, and is then re-fed through the scanner. Because the machine may have partially counted the ballot before the jam, the re-scan can result in a double count for that ballot, causing the poll tape total to exceed the number of ballots cast. Ballot jam re-scans are a documented, legitimate source of small discrepancies between poll tape values and ballot counts. A confirmed re-scan resolves the discrepancy to a match, with the re-scan recorded in its lineage. Documented in the 2024 General Election Wake County cases.
Ballot marking device (BMD). A voting machine that assists voters in marking paper ballots, which are then tabulated by separate equipment. Distinguished from direct-recording electronic (DRE) machines and optical scanners.
C
Canvass. The official post-election process in which a jurisdiction reviews and reconciles election data before certifying the results. Activities typically include reconciling the number of voters who checked in against the number of ballots cast, verifying that all precincts have reported, and reviewing provisional ballots. The canvass is the primary window for catching reporting errors through internal checks. AV’s before-certification window overlaps with the canvass period. Learn more here.
Cast vote record (CVR). A digital record, created by a voting machine, of how each ballot was interpreted. CVR review is a counting-layer verification method — it checks whether the scanner correctly read each mark. Distinct from AV, which checks the reporting layer. Learn more here.
Central count facility. An America Counts term for the facility where a jurisdiction centrally tabulates ballots — early and mail ballots, or all ballots under a central-count system. Also called a county counting center or central counting place. Where central tabulation happens, a precinct-level poll tape may not exist; the results or certificate may instead be posted or observable here. A central posting location. Learn more here.
Central posting location. An America Counts term for a jurisdiction-level place that is not itself a voting location but where poll tapes or precinct results may be posted or obtained: the county courthouse, the central count facility, or the elections office. One of the two families of poll tape posting location, distinguished from a voting location, where ballots are actually cast. Learn more here.
Central tabulation (central count). The process by which precinct-level vote data — carried on memory cards or transmitted electronically — is aggregated at a central facility (typically the county elections office) into jurisdiction-wide totals. Central tabulation is the point in the reporting layer where data from many sources converges into one system, making it a critical locus of potential reporting-layer failure. The Butler County 2008 GEMS sharing violation and the Baldwin County 2002 episode are both central-tabulation failures. Learn more here.
Certification. The official act by which a jurisdiction declares election results final. Certification is the dividing line between the before-certification and after-certification analysis windows. Once results are certified, changing them requires legal proceedings (recounts, election contests). Learn more here.
Chain of custody. The documented trail of how evidence (in AV’s context, poll tape recordings) was created, handled, stored, and analyzed. AV’s chain of custody includes GPS coordinates, timestamps, device metadata, vetting records, and archive publication. Learn more here.
Citizen auditor. An aspirational role category for Actual Vote users who go beyond recording poll tapes to participate in the full Comparison Analysis pipeline — transcription, comparison, investigation, and resolution. The term reflects AC’s long-term vision of widespread civic participation in vote reporting verification. Learn more here.
Citizens Audit Broward (CAB). A local civic group in Broward County, Florida, that operated as an audit committee in 2018 and 2020. The canonical example of an audit committee in this manual. Learn more here.
Comparison. Taking one number from the official results and comparing it against the corresponding vote total on a poll tape — usually a value transcribed from an Actual Vote video submission. Sometimes one official number corresponds to a sum of totals across several poll tapes: when two tabulating machines split the work for a single voting location, we typically get one tape from each, while the official results report that location as a whole — so to compare, we first add the corresponding numbers from the two tapes. On first pass a comparison is an exact match or an initial non-match; most are matches, and when one isn’t, we investigate further (see comparison outcomes). Learn more here.
Comparison Analysis. The end-to-end pipeline work that turns submissions into findings. The formal name for the analytical process at the heart of Actual Vote. Distinct from “Actual Vote analysis,” which is the larger scope (one election, one state, all the work). Learn more here.
Comparison-eligible transcribed value. Any vote total we manage to transcribe from a poll tape in an Actual Vote submission — every transcribed total is, in principle, eligible to be compared against the official results. An excellent recording makes every total readable and transcribable, and each becomes comparison-eligible once transcribed; a total that is too blurry, covered by tape, or hidden because the tape bent in the wind at the wrong moment can’t be transcribed, and so can’t be comparison-eligible. Whether an eligible value then becomes a possible comparison depends on how the official results are reported — some or all eligible values (possibly after aggregation) pair with official values to form possible comparisons, but not always: a state may withhold totals too small to preserve voter anonymity (as Georgia, Florida, and North Carolina do), or report results sorted differently than the poll tapes present them (see Vote Reporting in Georgia). Learn more here.
Comparison grain. The set of dimensions at which a specific Actual Vote analysis performs comparisons. The typical grain is County × Precinct × Voting Method × Contest × Choice. The grain may differ for specific jurisdictions (e.g., Georgia Early Voting uses vote center, not precinct). Learn more here.
Comparison logic. The set of mappings, aggregation rules, and edge case rules that determines how transcribed values are matched against official results for a specific Actual Vote analysis. Learn more here.
Comparison outcomes. The consolidated map of what can happen to one comparison. At first pass it is an exact match or an initial non-match. Review triages each initial non-match: AC’s own errors are corrected to a match (unreported); the rest are discrepancies, each described by magnitude, contest-flipping status, and an explanatory state that moves likely attributable → apparent → escalatable → escalated, exiting to a match whenever explained or terminating as a true discrepancy. Replaces the former “discrepancy hierarchy.” Learn more here.
Comparison status. The current lifecycle state of a comparison against a given official-results version: exact match · initial non-match · likely attributable discrepancy · apparent discrepancy · escalatable discrepancy · escalated discrepancy · true discrepancy · match · match (new official results) · official results ruled correct · unresolved (no court case) · uncharted territory. Updated in real time; the ordinary happy terminus is match. Tracked separately from lineage. Learn more here.
Contest-flipping / non-contest-flipping discrepancy. A discrepancy is contest-flipping when its resolution could change the true winner of the contest — the contest’s unresolved difference measured against the margin of victory, with direction. The complement, non-contest-flipping, is the default and is assumed unless noted. Orthogonal to magnitude: a thin race can be flipped by a small discrepancy, a comfortable one survives a large one. Judged per contest (not per comparison). Every discrepancy AC has found to date is non-contest-flipping; a confirmed contest-flipping discrepancy would be the most significant result in the project’s history and would take priority over all other work. Replaces the former “material / immaterial discrepancy” pair. Learn more here.
Context submission. An approved submission that depicts something other than standard transcribable poll tape content. Examples: a poll tape posted incorrectly behind glass, a polling place with no tapes posted at all, an unusual operational situation. Context submissions document conditions on the ground even though they don’t feed the comparison pipeline. Learn more here.
Counting layer. The portion of the election process that runs from ballot through scanner to poll tape — the process by which voter intent is converted into numerical totals by a voting machine. Distinguished from the reporting layer. AV does not audit the counting layer. Learn more here.
County courthouse. The county’s courthouse, which in some states is where every precinct’s results must be posted centrally soon after polls close. Alabama, for example, requires each voting place’s results to be posted at the county courthouse within two hours of poll closing (Ala. Code § 17-12-8), putting an entire county’s results in one place. A central posting location — valuable precisely because it aggregates a whole county’s tapes at a single address. Learn more here.
CVR. See: cast vote record.
D
Democracy Counts, Inc. The 501(c)(3) parent legal entity of America Counts. Founder and CEO: Dan Wolf. Corporate website: democracycounts.org. Learn more here.
Direct-recording electronic (DRE). A voting machine that records votes directly onto an electronic medium, typically via a touchscreen, without producing a separate voter-marked paper ballot. Some DREs include a voter-verifiable paper audit trail (VVPAT). DREs are relevant to AV because they produce poll tapes like other voting equipment, but the counting-layer risks differ from optical scanners. Learn more here.
Discrepancy. A real difference between the aggregated transcribed value and the official value — a difference in the reporting record, not an artifact of AC’s own process (which would be an initial non-match). Described along three independent axes: magnitude, whether it is contest-flipping, and an explanatory state (likely attributable vs apparent, and onward). All discrepancies are worth resolving; magnitude and contest impact set urgency, never whether it matters. Learn more here.
Discrepancy state. A per-analysis headline summarizing where its discrepancies stand: none_found · all_resolved · resolutions_pending · true_discrepancy_existed_but_resolved · true_discrepancy_exists. The first three are everyday badges; the last two — plus any unresolved apparent, escalatable, or escalated discrepancy — route to a special report design. A true discrepancy that ever existed takes precedence over all_resolved. Learn more here.
Discrepancy hierarchy. See: comparison outcomes (this taxonomy was reorganized; “hierarchy” is retired).
Dominion Voting Systems. One of the three major US voting system vendors alongside ES&S and Hart InterCivic. Dominion acquired the GEMS software line from Premier Election Solutions (formerly Diebold Election Systems). Referenced throughout the case studies.
E
Early Voting. Votes cast in person at a designated location before the official election day. Available in most US jurisdictions. One of the voting method categories in the comparison grain. In some jurisdictions (notably Georgia), Early Voting is reported at the vote center level rather than the precinct level, requiring comparison grain adaptation. (The voting method; for the physical place, see early voting site.)
Early voting site. A voting location open during the early-voting period before Election Day. Whether an early voting site posts a tape varies: some early-voting equipment prints nothing until the early-voting period closes, and many jurisdictions carry early ballots to a central count facility to be tabulated instead. Learn more here.
Election Day. The official day on which an election takes place. Votes cast in person on Election Day are the most common voting method historically and the most common subject of AV poll tape recordings.
Election Management System (EMS). The software and hardware used by elections offices to program voting machines, aggregate results, and produce official reports. The EMS is the system that produces the official results against which AV compares its transcribed poll tape values. Learn more here.
Election Night Reporting (ENR). The system by which elections offices publish preliminary results on election night, typically through a public website. ENR data can be scraped as a source of official comparison values (Method 5 special cases).
Election-day voting site. An America Counts term for a voting location open on Election Day — either a precinct-based polling place or an election-day vote center. AC uses this term when describing the places a volunteer can go on election night to record posted tapes. The counterpart to an early voting site. Learn more here.
Election-jurisdiction pair. One distinct (election × jurisdiction) unit of audit coverage — a single election in a single jurisdiction, most often one county in one election. Cumulative totals sum these across every report, so a report covering three counties in one election contributes three election-jurisdiction pairs. Learn more here.
Elections office. The generic term in this manual for the government office in a jurisdiction that administers elections. The specific name varies by state: Supervisor of Elections (FL), Board of Elections (NC), County Clerk (many states), Registrar of Voters (CA), Municipal Clerk (WI). Often abbreviated SOE in informal AC usage. In the place taxonomy it is a central posting location, where results may be posted or obtained by appointment or records request. State-specific terminology is cataloged in B.3.
ES&S (Election Systems & Software). The largest US voting system vendor by market share. ES&S equipment is referenced in multiple case studies and its ExpressVote system is the subject of specific poll tape format notes. See also: Dominion Voting Systems, Hart InterCivic.
Escalatable discrepancy. An apparent discrepancy that neither AC’s own work nor a cooperative ask to the SOE could resolve, reaching the point where it must be escalated to a party able to bring a court case. AC escalates only what it cannot explain. Also called an unresolved discrepancy. It may still resolve before any handoff — a belated SOE explanation returns it to a match, a belated SOE acknowledgment makes it a true discrepancy — or the court case may fail to materialize, leaving it unresolved. Learn more here.
Escalated discrepancy. An escalatable discrepancy now being litigated by a court-capable party — a decision point whose outcome depends on the court. Known branches: the court finds AV’s evidence correct and the official results are restated (→ a true discrepancy in the version that contained it, then a match against the new official results); the court rules the official results correct; or uncharted territory. “Escalation” applies only here, at the court stage. Learn more here.
Escalation pathway. A procedural route through which AV evidence can reach decision-makers or enforcement bodies when a discrepancy is not resolved through cooperative contact with the elections office. Six pathways are identified: regulatory complaints, media and public transparency, supporting candidate action, public records litigation, federal civil rights litigation, and criminal referral. Learn more here.
Evidence preservation. The structural function of the Actual Vote Video Archive as a permanent, timestamped, publicly accessible record of poll tape data that exists independently of any government office’s record-keeping. Evidence preservation becomes critical when original election records are destroyed — as in Ohio 2004, where county records were destroyed in violation of a federal court order, and the 1948 Texas Senate race, where Box 13 evidence disappeared entirely. AV’s archive ensures that once a poll tape is recorded and approved, the data cannot be lost through negligence, policy, or intent.
Exact match. A comparison whose aggregated transcribed value equals the corresponding official value exactly on first pass. The most common comparison outcome, and one kind of match.
F
FOIA (Freedom of Information Act). The federal Freedom of Information Act and its state-level equivalents (e.g., Florida Sunshine Law, NC Public Records Law, TX Public Information Act). Used in Method 4 access for obtaining poll tape images from elections offices. Learn more here. Templates in B.12.
Fully verified contest. A contest in which an Actual Vote analysis audited every officially reported vote — zero unaudited values at the contest’s reporting grain. Errors in vote reporting can only hide in votes an audit did not check, so a fully verified contest’s outcome is independently confirmed outright, subject only to the matched poll tape caveat. First achieved in the NC 2024 Primary Report, where complete audits of Alexander and McDowell counties fully verified five contests. See also: overstatement-protected contest.
Fraction Magic. A proof-of-concept attack on GEMS tabulator software, developed by Bennie Smith, demonstrating that votes could be weighted as fractions rather than whole numbers, enabling undetectable manipulation. A counting-layer vulnerability — if applied before the poll tape prints, AV would not detect it. Full treatment in the Fraction Magic case study and the incident taxonomy.
G
GEMS (Global Election Management System). Election management software originally developed by Diebold Election Systems (later Premier Election Solutions, then acquired by ES&S, then Dominion). GEMS is the tabulation software at the center of several case studies, including the sharing violation bug documented in Butler County 2008 and the Shelby County 2015 discrepancy. See also the Fraction Magic case study.
Grain. See: comparison grain.
H
Hart InterCivic. One of the three major US voting system vendors alongside ES&S and Dominion Voting Systems. Hart equipment is referenced in the case studies. See also: ES&S, Dominion Voting Systems.
I
Initial non-match. Any comparison that does not match on the first pass, for any reason — the raw funnel that review then triages. Review assigns each a hypothesis where one exists: some are AC’s own errors (transcription, matching, mapping) and resolve trivially to a match (and are not reported), the rest become discrepancies. “Non-match,” not “mismatch,” keeps it neutral; “initial” marks the pre-review state. Learn more here.
Invisible spread. A pattern in which reporting-layer errors are distributed across many precincts at magnitudes too small to trigger aggregate-level detection, but which would be caught by systematic precinct-level comparison. The concept is illustrated by the Florida 2000 case study, where the statewide margin was 537 votes across 5,963 precincts — a spread that would be invisible in county-level totals but detectable if precinct-level poll tape data were compared against precinct-level reported results. Invisible spread is a core argument for AV’s precinct-level methodology.
L
L&A testing (Logic and Accuracy testing). Pre-election testing in which election officials run a known set of test ballots through voting equipment and verify that the machine produces the expected results. L&A testing checks counting-layer accuracy before the election. It does not check the reporting layer. Learn more here.
Likely attributable discrepancy. A discrepancy for which AC holds a working hypothesis about the vote-reporting logic that would resolve it — a hypothesis of convenience, not a verdict. It is “likely” because we expect routine confirmation (by AC’s own check — a write-in equivalence check, a ballot-jam re-scan — or by a lighter ask to the SOE), but the resolution depends entirely on the evidence. Exits: confirmed → match; a plausible explanation accepted for a non-contest-flipping, benign-magnitude difference without full confirmation → match (plausible explanation); no satisfactory resolution → apparent (and only from there, if it proves a real error, onward to a true discrepancy — a likely-attributable discrepancy never becomes one directly). It is a transient, in-flight state: summary statistics report its resolved forms (matches, or apparent discrepancies), not the in-flight state itself, so it is not a stand-alone stat line. Any magnitude; can be contest-flipping. Learn more here.
Lineage. The story of how a comparison reached its current status — orthogonal to that status. Ranges from trivial (always an exact match, or an own-error fix) to maximal (was litigated, then restated to a match). A match with non-trivial lineage is the interesting case to write up; a match’s lineage, not a separate status, is where AC records a noteworthy history. Learn more here.
Lone user. A category of Actual Vote user: a single individual who uses the app to record poll tapes on election night, typically working alone. The most common entry point for new users. Also informally called “lone wolf.” Learn more here.
M
Magnitude. The size of a discrepancy, reported as two numbers: the absolute magnitude (the votes in question — the delta between transcribed and official values) and the relative magnitude (that delta as a share of a stated denominator, normally the contest total at the comparison grain). Both matter, differently: 10 votes is large relative magnitude in a 20-vote contest, small in a 2,000-vote one. AC reports the figures rather than a fixed low/high label, keeping every claim falsifiable. Magnitude governs the urgency and effort AC applies, proportionally — never whether a discrepancy is worth resolving. Orthogonal to contest-flipping status. Learn more here.
Match. A comparison whose value agrees with the poll tape as reflected in the official record — either by first-pass equality (an exact match) or because a discrepancy was explained and reconciled. Most matches are exact; where a difference was reconciled, AC names two subtypes by the strength of the explanation — a plausibly equivalent write-in situation (arithmetically proven) and a match (plausible explanation) (a plausible, non-contest-flipping explanation accepted without full confirmation). It carries a lineage (trivial → maximal). Match is always relative to the poll tape. Learn more here.
Match (plausible explanation). A match reached when a likely attributable discrepancy has a plausible benign explanation — from AC’s own reasoning, an ask to the SOE, or the official record — that AC accepts without full per-value confirmation, because the difference is non-contest-flipping and of a magnitude the analyst judges benign (ideally within what the cited mechanism can produce). It is the weaker sibling of the arithmetically proven PEWS match: the explanation is credible, not confirmed. Each is itemized in the report with its magnitude, mechanism, and the analyst’s recorded rationale, and stays revisable — the door is never closed. Consistent with proportionality (see the reasoning that governs spot-checking): AC does not spend its own or an elections office’s finite time chasing full confirmation of a low-stakes difference that already has a credible account. It counts as a match, never a discrepancy, in summary statistics. Internally nicknamed a “match modulo plausible explanation” (MMPE). Learn more here.
Match (new official results). A match reached against a restated official-results version — which by definition means a true discrepancy existed in the prior version. A good current status carrying maximal lineage; always routes to a special report design. Held distinct from the ordinary match because the agreement is against a different version of the official record. Learn more here.
Matched poll tape caveat. AV’s most important stated limitation: a match between a transcribed poll tape value and an official reported value confirms that the reporting layer was accurate, but says nothing about whether the poll tape itself is correct. If a voting machine miscounted ballots, the poll tape would reflect the miscount, and AV would report a match — because the reported value matches the poll tape value, even though both are wrong. AV verifies vote reporting, not vote counting. This caveat is a standing commitment to intellectual honesty. Learn more here.
Memory card (memory pack). The removable physical media (typically a compact flash card or proprietary cartridge) used to transport vote data from a precinct tabulator to the central tabulation facility. After polls close, poll workers remove the memory card from the scanner and deliver it — by hand or courier — to the county elections office for upload into the Election Management System. Memory cards are the physical link between counting and reporting layers, and their handling is implicated in several case studies: the Butler County 2008 GEMS sharing violation, the Baldwin County 2002 discrepancies, and Florida 2000. Synonyms vary by vendor and jurisdiction.
Method 1, 2, 3, 4, 5. The five access methods for obtaining poll tape recordings. See: access method. Learn more here.
N
NC 2024 Primary Report. The canonical published example of an Actual Vote analysis report. Covers seven North Carolina counties, the March 5, 2024 primary election, with 52 transcribable submissions, 321,502 votes audited, and zero true discrepancies. Reproduced as B.6.
O
Official results. The vote totals published by the responsible elections authority — the values against which AV compares transcribed poll tape values. Sourced per analysis from the authority’s published data (e.g., the North Carolina State Board of Elections results files). The “official” qualifier identifies the source, not a presumption of accuracy; independently checking their accuracy at the reporting layer is the purpose of the comparison.
Official results ruled correct. A rare court terminal in which a court rules the official results correct because AV’s poll-tape evidence fails to persuade the court, despite the demonstrated difference. Neither a match nor a true discrepancy; surprising to us, but a possible outcome of litigation. Routes to a special report design. Learn more here.
Official-results version. A dated snapshot of a jurisdiction’s reported results — v1 unofficial (election-night) → v2 amended → v3 certified → (rarely) v4 post-court. AV evaluates comparisons against each version, so comparison status and true-discrepancy status are relative to a version. Learn more here.
Optical scanner. A voting machine that reads marks on hand-marked or BMD-marked paper ballots and tabulates the results. The most common type of vote-counting equipment in use in the United States. Optical scanners produce poll tapes, which are the primary evidence AV captures.
Overstatement-protected contest. A contest whose unaudited reported votes number fewer than its margin of victory. Even if every unaudited reported vote were fictitious, the audited (verified) votes alone would preserve the winner’s lead, so overstatement errors in vote reporting cannot have changed the outcome. The protection bounds fictitious reported votes only; votes genuinely cast but never reported at all are the province of turnout reconciliation rather than this comparison. A fully verified contest is the limiting case, with zero unaudited votes.
P
Parallel testing. Election-day testing in which specially marked test ballots are fed through active voting machines at select polling places to verify real-time counting accuracy. A counting-layer check. Learn more here.
Plausibly equivalent write-in situation (PEWS). A comparison difference that arises because the poll tape and the official results represent the same write-in votes two different ways, rather than because anything went wrong. The tape prints one catch-all raw write-in count; during canvass those write-ins are resolved — credited to a listed candidate the voter wrote in, credited to a qualified write-in candidate listed separately in the official results, or discarded as invalid — so the official representation legitimately differs from the tape’s. One resolution event can surface as several first-pass mismatches in the same contest (a listed candidate +1, the write-in line −1). A situation is declared plausibly equivalent only when the write-in equivalence check passes; a confirmed PEWS resolves the discrepancy to a match (recorded in its lineage), the same treatment as a confirmed ballot jam re-scan. “Plausibly” is deliberate: the check verifies that the two representations are arithmetically equivalent accounts of the same votes, not that the canvass records themselves were inspected. Learn more here.
Poll observer. A person formally appointed to observe election proceedings, typically by a political party or candidate. Also called poll watcher, election observer, or other state-specific names. Relevant to Method 2 access. State-specific terminology in B.3.
Poll tape. A strip of paper printed by a voting machine at the close of polls, showing the totals tabulated by that machine. The atomic data object that Actual Vote captures. Contents typically include: machine ID, precinct or vote center ID, election date, contest names, candidate names, vote totals, and sometimes timestamps and poll worker signatures. Types include results tapes (the subject of AV analysis), zero tapes (printed before polls open to confirm zero starting counts), and test tapes (printed during pre-election testing). Learn more here.
Poll tape posting location. An America Counts term for any physical place where a poll tape — a machine results printout or signed certificate of results — can be posted for public view or otherwise obtained. It covers two families: voting locations and central posting locations. This is the organizing idea behind AC’s state-by-state posting research: where a tape can be found is what determines which access method a volunteer uses. Learn more here.
Poll worker. A person formally appointed to staff a polling place during elections, typically by the local elections office. Also called election inspector (WI), officer of election (VA), precinct election official (OH), or election judge. Relevant to Method 2 access. State-specific terminology in B.3.
Polling place. A physical place where voters cast in-person ballots under the supervision of election workers during set hours. In the traditional model it is precinct-based: voters are assigned to one polling place by where they live, and a single polling place may serve one precinct or several. A kind of voting location. (Avoid “polling station,” which is sometimes used to mean the room rather than the building.) Learn more here.
Possible comparison. A comparison between the official results and the transcribed poll-tape totals that we can actually carry out. Depending on what poll-tape data we could transcribe and how the official results are reported, not every contest-choice in the official results can be matched to tape totals — so for any Actual Vote analysis, the number of possible comparisons is at most the number of comparison-eligible transcribed values. Learn more here.
Posting compliance. The rate at which polling places comply with statutory or regulatory requirements to publicly display poll tapes after polls close. AV’s theory of change depends on poll tapes being physically accessible; low posting compliance directly limits the number of recordings that can be obtained via Method 1 (election night posting). The Broward County 2018 analysis documented approximately one-third non-compliance — one in three polling places either failed to post tapes or posted them in ways that made recording impractical. Posting compliance is a measurable indicator of election administration quality and a key operational factor in planning AV recording operations. Learn more here.
Precinct. An election-administration division: a geographic area that determines which contests its residents may vote on, and which serves as a unit for tabulating and reporting results. A precinct is not a physical place — it is mapped onto a voting location. A polling place may serve one precinct or several, and under vote centers precincts are decoupled from the site entirely. One dimension of the standard comparison grain. Learn more here.
R
Recount. A post-election procedure in which ballots are re-tabulated, either by machine or by hand. Recounts check counting-layer accuracy — whether machines read ballots correctly. They do not check reporting-layer accuracy unless the recount also compares retabulated precinct totals against originally reported values. Learn more here.
Red shift. A pattern in which official vote counts diverge from pre-election polls or exit polls in a direction favoring more conservative candidates. The term is used in election forensics. AV does not measure or claim to detect red shift; AV checks reporting-layer accuracy at the precinct level regardless of which direction any discrepancy runs.
Rejected submission. A submission that failed AC’s vetting process. Visible only to the submitter. Common rejection reasons include PII captured in the recording, no usable poll tape content, and wrong content type. Learn more here.
Report (Actual Vote report). The published output of one Actual Vote analysis: a single document presenting the comparisons, investigations, and findings for one election in one jurisdiction (or, in the compiled report, a growing set of smaller submissions). “Reports published” counts the reports whose pages are live on this site. Learn more here.
Reporting layer. The portion of the election process that runs from poll tape through all subsequent steps to published official results — the process by which the voting machine’s output is transmitted, aggregated, and certified into the numbers the public sees. Distinguished from the counting layer. AV audits the reporting layer. Learn more here.
Results tape (results poll tape). The kind of poll tape that Actual Vote analyzes: the strip a voting machine prints at the close of polls showing the vote totals it tabulated. Distinguished from a zero tape (printed before polls open to confirm every count starts at zero) and a test tape (printed during pre-election testing). When this manual says “poll tape” without qualification, it usually means a results tape. Learn more here.
Risk-limiting audit (RLA). A post-election audit that examines an escalating random sample of ballots until statistical confidence is reached that the reported winner is the actual winner. An RLA is defined by its risk limit — a statistical guarantee that depends on random sampling. RLAs check counting-layer accuracy. They do not check reporting-layer accuracy — the Prince William County 2020 case demonstrated that a RLA can confirm counting with 99%+ confidence while a 4,000-vote reporting error goes undetected. Actual Vote is not a risk-limiting audit: AV performs the same batch-comparison check at full coverage of the audited precincts’ reporting chain and computes no risk limit — for the audited precincts the comparison is exhaustive rather than sampled, so none is needed. Learn more here.
Rolling-log report. A live-updated page maintained during an active analysis, capturing findings and observations in reverse-chronological order with timestamps. The 2024 General Election News Page is the canonical example. Contrasted with the static analysis report. Learn more here.
S
Scrutineers. A partner organization that helps connect volunteers with AC and supports poll tape recording efforts across multiple jurisdictions. Learn more here.
Seven-check investigation process. The structured set of checks AC applies to a non-match — transcription accuracy, matching logic, aggregation rules, missing or duplicate tapes, undocumented vote reporting logic, operational context, and independent re-verification. The checks triage an initial non-match (catching AC’s own errors) and inform AC’s investigation of a discrepancy before any heavier ask to the SOE — not a pre-court gate. Learn more here.
Sharing violation. A software bug in GEMS tabulation software in which simultaneous memory card uploads cause one card’s data to overwrite another’s without generating an error message. The canonical example of a silent reporting-layer failure. Documented in Butler County 2008 and implicated in Shelby County 2015. Full treatment in the incident taxonomy, Category 2.
Stated assumption. A documented inference about how a jurisdiction reports votes — an aggregation grain, an identifier reconciliation, the treatment of unpublished values — that AC adopts because the public record does not fully specify the reporting process and exhaustively querying the elections office would be inappropriate. A stated assumption is distinguished from a hidden one by being recorded in the analysis manifest and methodology notes, so a reader knows exactly which numbers rest on documented fact and which on reasoned inference. AC’s standard for adopting one is over-determination: the inference must be forced by many independent observations at once, not merely consistent with them. Contrast with a confirmed fact (documented or county-verified). Learn more here.
SOE (Supervisor of Elections). The Florida-specific title for the county-level elections administrator. Used informally by AC as a generic abbreviation for any elections office. See: elections office.
Static analysis report. A single document published once an analysis is complete, describing what AC did and what AC found in a structured, definitive format. The NC 2024 Primary Report is the canonical example. Contrasted with the rolling-log report. Learn more here.
Structural gap. The observation, central to this manual’s argument, that nearly every established election integrity safeguard checks counting-layer accuracy while almost nothing systematically checks reporting-layer accuracy. AV fills this gap. Learn more here.
Submission. One video uploaded (submitted) to the Actual Vote Video Archive from the Actual Vote app. Most submissions depict a single results poll tape, but users are also encouraged to submit videos of other vote-reporting subjects — zero tapes, poll-closing procedure (with permission), or voting locations where poll tapes are supposed to be posted publicly but don’t seem to be. When someone can’t use the app to record poll tapes — they’re forbidden from doing so, their device won’t run the app, or it otherwise fails them — they may instead submit photos or videos taken with their device’s own camera app, of any of those same subjects, directly to America Counts (by email, Dropbox, Google Drive, and the like). These still count as Actual Vote submissions even though they don’t come from the app itself; they won’t appear in the Archive until America Counts uploads them there. Submissions made through the app carry associated metadata (GPS coordinates, timestamp, device information, and a user-entered precinct identifier). Each submission has a unique integer ID and may be classified as approved, rejected, practice, transcribable, not-practical-to-transcribe, context, or analyzed. Learn more here.
T
Transcribable submission. A submission that shows at least one poll-tape vote total we can transcribe. Most of the time those transcribed totals can be used in a comparison against the official results — but not always (see Vote Reporting in Georgia). More formally: an approved submission depicting at least part of one results poll tape, with state, county, election date, voting method, and precinct clearly readable in the header, and containing at least one legible contest/choice/vote-total triple. The submission type that feeds the Comparison Analysis pipeline. Learn more here.
Transcribed value. A single numeric value read from a poll tape and entered into the structured transcription dataset. Each value is associated with a submission ID, contest, choice, voting method, and precinct. Learn more here.
Transcription. The process of reading values from a transcribable submission’s poll tape into the structured transcription dataset. Currently done by AI-assisted human-in-the-loop workflow. Learn more here.
True discrepancy. A discrepancy confirmed by the responsible authority — the elections office acknowledging the error, or a court — to be a genuine vote-reporting error in a specific official-results version. AC never self-declares one; both confirmation routes converge here. Once the official results are restated, the comparison is a match against the new official results; a true discrepancy that is never restated stands. Version-relative, so the headline measure is true discrepancies that ever existed in this analysis. To date, zero. Learn more here.
U
Uncharted territory. A placeholder terminal acknowledging that AC has never litigated an escalated discrepancy, so the full set of court outcomes is genuinely unknown. Present so the model does not overstate its own completeness. Learn more here.
Unresolved discrepancy. See: escalatable discrepancy.
Unresolved, no court case. A terminal in which an escalatable discrepancy never reaches a court — the case fails to materialize for any of many reasons (no party will bring it, standing, cost, timing) — so the discrepancy simply stands, unadjudicated. Neither a match nor a true discrepancy. Routes to a special report design. Learn more here.
User Agreement. The canonical AC document that governs the relationship between Actual Vote users and AC. Reproduced in B.4. Cross-referenced from the Access Methods chapter.
V
Valid comparison. Retired term — see comparison. A comparison that can be made is valid by definition, so America Counts now writes simply “comparison.” Where it matters that some transcribed values had no official counterpart to compare against, use “possible comparison” (the ones we could actually make) versus “comparison-eligible transcribed value” (everything we transcribed).
Vetting. The human review process through which AC evaluates each submission for quality, privacy, completeness, and usability before approving it for public display in the archive. Vetting criteria include legibility, correct poll tape type, no PII visible, and metadata consistency. Learn more here.
Video-recording team. A category of Actual Vote user: an organized group that plans and executes poll tape recording operations across multiple polling places. Also informally called a “wolf pack.” Learn more here.
Vote center. A voting location where any registered voter in the jurisdiction may cast a ballot, regardless of which precinct they live in — an alternative to neighborhood, precinct-based polling places. Because a vote center must serve every voter in the jurisdiction, every contest anywhere in that jurisdiction can appear on someone’s ballot there — and therefore every one of those contests prints on the vote center’s poll tape. This can stack many blocks of purely local contests (ones that apply to only a single city or district) onto a single tape, producing very long poll tapes; the California 2020 analysis is a good example. When results are reported at the vote center level rather than the precinct level, the comparison grain must be adapted accordingly. Learn more here.
Vote reporting. The process by which individual polling place results (as printed on poll tapes) are aggregated and published as official election results. Distinguished from vote counting, which is the process by which voting machines tabulate individual ballots. AV verifies vote reporting, not vote counting. The distinction is central to understanding AV’s scope. Learn more here.
Votes audited. The headline size measure of an Actual Vote analysis: the sum of all transcribed vote totals whose reporting was independently checked — vote totals read from poll tapes and compared against the corresponding official results. “Audited” here is narrow and specific: it means only that the reporting of these votes was verified with the Actual Vote system — do the published totals match what the machines printed? It is not a complete audit of every election-security layer; in particular it does not check the casting, counting, or tabulation of the ballots themselves (the counting layer). 321,502 in the NC 2024 Primary Report.
Voting location. An America Counts term for any place where a person votes in person, whether on Election Day or during early voting — i.e. an election-day voting site (a polling place or vote center) or an early voting site. It excludes drop boxes and mail-ballot return, where no in-person voting happens and no tape is produced on site. AC uses “voting location” as the umbrella above polling place and vote center because neither of those is neutral: “polling place” implies the precinct-based model, and “vote center” is one particular type. One of the two families of poll tape posting location. Learn more here.
Voting method. A category of how votes were cast: Election Day, Early Voting, Absentee/Mail, Provisional, etc. One dimension of the standard comparison grain. Specific voting methods and their labels vary by jurisdiction.
Voting station (voting booth). The spot within a voting location where an individual voter actually marks or records their ballot — the booth or enclosure. Distinct from the voting location itself, which is the building or site.
W
Wanna Vote. A separate app from America Counts (in development) for vote counting verification — a distinct problem from vote reporting verification. Wanna Vote uses citizen canvassing surveys to build an independent ground-truth dataset of how people voted, then compares that to official results. AV and Wanna Vote together would cover both halves of the verification problem (reporting and counting), but they are methodologically distinct. This manual is about Actual Vote; Wanna Vote is mentioned only to clarify the boundary. Learn more here.
Wolf pack. Informal term for an organized local team conducting poll tape recording operations. See: video-recording team.
Write-in equivalence check. The five-condition arithmetic test for a plausibly equivalent write-in situation. For a contest, with diff = official − tape per listed candidate: (1) no listed candidate loses votes (every diff ≥ 0); (2) redistributed = the sum of positive diffs (write-ins credited to listed candidates); (3) invalid = tape write-ins − redistributed − all official write-in lines (including qualified write-in candidates) must be ≥ 0 — you cannot drop more write-ins than the tape recorded; (4) the totals reconcile exactly: tape total − invalid = official total; (5) the tape is internally consistent (its choice values plus overvotes and undervotes reproduce its own printed Total Votes line). All five must hold, and the magnitudes are inherently bounded by the tape’s raw write-in count. A passing contest collapses its first-pass mismatches into one review item; a human confirms before any comparison is reclassified as a match. In jurisdictions that publish no write-in lines at the precinct level (Georgia), the raw write-in count generalizes to the tape’s full non-certifiable pool — the catch-all Write-In line plus any lines for court-disqualified candidates whose votes are void — with the same five conditions applied. Learn more here.