David Crowley won Wisconsin's Democratic governor primary by less than one percentage point. Abdul El-Sayed's victory in Michigan's Democratic Senate primary came by roughly 15,000 votes—also about a point—according to Associated Press reporting.
Those results support a modest conclusion: two important Midwestern Democratic electorates were closely divided. They do not prove that the same factions are equally strong nationwide, that one campaign tactic caused either outcome or that the November electorate will split the same way.
Close-election coverage often goes wrong before interpretation begins. A newsroom sorts a margin column, labels the first ten records the “closest races” and publishes. That procedure can compare incomparable things.
A margin needs a denominator
A 500-vote lead can be enormous in a contest with 1,000 votes and negligible in a contest with 1 million. Vote margin and percentage-point margin answer different questions. Both should be shown.
Uncontested races must be excluded. So must records in which a winner is marked despite a zero-vote placeholder. Incomplete reporting can make a temporary margin look final. Duplicate contest rows can count the same race twice.
The BohoNews research table contains examples of these hazards, including negative calculated margins where a provider's winner marker does not match a simple vote-order calculation. Those rows are diagnostic evidence of a normalization problem, not publishable evidence of an upset.
Election systems determine the correct round
In an ordinary plurality contest, comparing the leading candidate with the runner-up may be straightforward. In ranked-choice voting, the meaningful result may be the final active round after lower-ranked candidates are eliminated and votes transfer. First-round totals answer a different question.
In a top-two primary, the important threshold may be who advances, not who becomes a party nominee. In a majority-runoff state, a first-place candidate can still fall short of nomination. Treating all of those outcomes as identical margins produces false precision.
A call, a count and a certification are different
The Associated Press can call a race before every ballot is counted when its analysis finds no path for the trailing candidate. Election officials publish unofficial totals, conduct a canvass, resolve eligible provisional and cured ballots, perform required audits and eventually certify.
Recount law adds another layer. Some states trigger a recount automatically below a threshold; others allow a candidate to request one and may assign costs depending on the outcome. A narrow AP-called race is not automatically in a recount, and a recount does not imply that the original count was improper.
What Wisconsin and Michigan do tell us
Wisconsin's result denies both Democratic factions an easy claim of dominance. Crowley won, but Hong's support was broad enough to make party unity a substantive task rather than a ceremonial statement.
Michigan's result shows that a progressive candidate can win a high-spending statewide primary in a genuine presidential battleground. The narrow margin also means a large share of Democratic primary voters preferred another choice. El-Sayed must persuade those voters to remain in coalition while reaching an electorate that did not participate in the primary.
Close races are valuable because they reveal coalition boundaries. They become misleading when decimals are used to imply certainty about causes.
The publishable standard
Before BohoNews publishes a national closest-races table, each included contest should have an official source, a defined final round, complete or certified totals, a verified election type and a documented denominator. The table should display both vote and percentage-point margins and state what was excluded.
Until those requirements are met, the responsible article is not a ranking. It is an explanation of why a ranking would be premature.
Methodology
This article deliberately does not publish an automated top-ten margin list. A quality review found ranked-choice, zero-vote, incomplete and winner-order edge cases in the research table. The Wisconsin and Michigan descriptions rely on attributed AP reporting rather than the uncorrected rank output.