Utah

Utah's aggregated canvass reports, per county per election, whether ballot totals reconcile against voter credit — and where they do not, the county's written explanation. Counties reporting a discrepancy: 4 at the 2024 presidential primary, then 6, 8, 9, 10, 12, and 17 at the 2024 general, out of 29. A consumer that treats a certified total as a single authoritative number has nowhere to put this, and one that reads the discrepancies as errors in the count misreads them: they are the difference between ballots counted and voter credit recorded, which is a different quantity.

The statewide canvass covers multi-county federal, statewide, legislative and State Board of Education offices only — 3 U.S. House districts, 7 senate districts and 17 house districts in 2026 — and states on its first page that county and local board of education results are held by the county clerk. So the state's certified record is incomplete by design, and the missing part is not a harvest gap but a jurisdictional boundary. A single-county legislative seat is certified nowhere in the state file.

The certification reports active registrations, ballots cast and turnout separately by party, and the Democratic columns are blank in 24 of 29 counties. Turnout is computed within each party against that party's own active registrations, so a county's Republican rate and another county's Democratic rate measure different populations and cannot be compared or averaged. Reading a single turnout figure off this document without carrying the party scope produces a number that describes nobody.

State legislature and state and local school board candidates filed January 2 to 8; U.S. House candidates March 9 to 13; unaffiliated candidates by June 15; write-ins by August 28. Congressional candidates could separately file intent to gather signatures from January 2 to March 13, so a candidate may pursue nomination by convention and by signature at once and the two tracks close on different dates. A single candidate-filing deadline does not exist here, and a candidate list captured between windows is incomplete for some offices and final for others.

Registration is reported across nine parties plus unaffiliated: Constitution, Democratic, Forward, Green, Independent American, Libertarian, Peoples' Freedom, Republican and Unaffiliated. Peoples' Freedom shows zero active, zero inactive and zero total — a ballot-qualified party with no registrants at all. A schema that infers the party vocabulary from non-empty registration counts will not know it exists, and one that treats a zero row as a data error will drop a real entity.

In the August 3 statistics the Active column totals 1,775,860 and the Inactive column totals 305,649 — both matching the state's own stated totals exactly. Their sum is 2,081,509, while the published Total is 2,081,525, sixteen higher; individual party rows differ by up to nine in both directions. The columns are each internally consistent, so the Total column appears to count a status the table does not display. Deriving a total by addition and comparing it against the published one will disagree, and neither figure is wrong so much as differently scoped.

Utah publishes its own definition: an inactive voter is one who has not voted in two regular general elections and has failed to respond to a notice from the county clerk. That is 305,649 people, 14.7% of the total roll. Because Utah votes by mail, the active-inactive line governs who a ballot is sent to, so the distinction is operational rather than clerical — and a turnout denominator that includes inactive registrants counts people the county did not mail.

Utah's election-night portal runs the same product as Virginia's, Idaho's and North Dakota's new one — the content security policy names the vendor directly. Virginia deploys it as a single statewide instance, Idaho as forty-four separate per-county instances, North Dakota with precinct display switched off entirely. So the feed's shape is a deployment choice by the customer and says nothing about the state's structure, and a recipe that works against one state's instance may address the wrong number of jurisdictions in the next.