How TPR will call election outcomes, and when a lead is not enough

What a TPR label means

Election reporting should answer the question readers are asking without disguising what remains unknown. TPR will distinguish a recorded lead, a mathematically secured ward outcome, a statistical projection, a completed dashboard allocation, a formal declaration and the election of a mayor. None is a substitute for the others.

The standard takes useful principles from the Associated Press and Decision Desk HQ: traceable official inputs, attention to outstanding returns, a separate decision process and a visible correction record. Their US race-calling practice cannot simply be copied onto South Africa’s mixture of ward and proportional representation contests.

Recorded lead means a party or candidate has the largest accepted total on a specified ballot at a specified time. It carries no claim about the eventual outcome. A district completion percentage is not a vote completion percentage and is not a random sample of the electorate.

When a ward can be called before every return

A mathematical ward call needs verified individual candidate identities, accepted ward-ballot totals and an authoritative upper bound on every valid ballot still able to enter that ward contest. The leader’s margin must exceed that full bound, including unresolved material corrections, outstanding special votes and other legally includable ballots. A guessed turnout or average station size cannot supply the bound.

A grouped INDEPENDENTS party row does not identify a candidate. If the feed lacks candidate-level results or the outstanding-ballot evidence, this route to a call is unavailable. The current TPR dashboard adapter does not by itself satisfy those requirements. No mathematical ward call is being issued by this article.

Statistical projections need a tested model

A TPR projection must name its contest and ballot, use a versioned model validated against completed official elections, account for the non-random arrival and geographic mix of returns, disclose uncertainty and pass stress tests for outstanding votes and material corrections. The model’s output and a documented editorial decision must both support the call.

This policy does not activate a projection model or manufacture a confidence threshold. TPR has not validated and deployed its own municipal forecasting model. A date-stamped external projection may be reported with the forecaster’s name, scope and limitations, clearly distinguished from TPR’s assessment and the IEC result.

The CSIR says its 2026 model uses historical voting patterns and clustering to account for how district results arrive. It describes its output as independent statistical estimates, with official declarations remaining the IEC’s responsibility. A second website repeating a forecast is not independent confirmation of that model.

What can be concluded from a completed allocation

For a council allocation, the full party and independent seat rows must reconcile to the expected council size. Ward and PR seats must add correctly for each contestant, and the current event, municipality, complete district and ward inventories and source date must be verified. Unresolved source revisions remain held for review.

A party with more than half the allocated seats has an outright majority in that reviewed allocation. If none does, the complete allocation has no outright majority. Being the largest party is not the same as winning that majority. TPR labels this “dashboard allocation” until it verifies the corresponding formal declaration.

Partial PR percentages cannot simply be multiplied by council size to produce a final seat forecast. Ward results, independent councillors, statutory allocation rules, ties and any applicable overhang provisions must be handled by a separately validated allocation system. Totals that add up are necessary, but not sufficient proof that a model implements those rules correctly.

Review, revision and public accountability

Every proposed call needs retained source bytes and hashes, source and retrieval times, geographic and ballot scope, the calculation or model version, the outstanding-vote assessment, material disputes, the reason for the decision and a recorded reviewer sign-off. Data checking and editorial judgment are separate steps. A claim without the required evidence stays uncalled.

TPR is not claiming that an independent human decision desk has reviewed a race when it has not. A statistical call requires a qualified independent review of the model and decision. Two copies of the same feed or two calculations by the same automated pipeline do not constitute that independence.

A changed call must be corrected prominently with the original claim, correction time, reason and effect. Earlier dated reports remain identifiable. An official declaration needs the dated IEC record itself; the name of a mayor requires the council’s recorded election or another authoritative appointment record.

Readers can ask a precise question through TPR’s evidence channel and request a correction from each article. The aim is an intelligible public record: what is known, how it is known and what evidence would change the conclusion. Speed is valuable when those distinctions survive it.

Send a specific source or question · Request a correction

AP race-call methodology · DDHQ official-data and projection policy · CSIR 2026 model explanation