Real-time leading-party board. Sorted by completion. Margins are leading-vote-share spreads, not seat counts. Refresh cadence — 90 seconds.
The seats where the winning margin is currently inside our 95% confidence envelope. These move first — and they move the alliance total most.
Every entry is a signed declaration from the Matrize control room, time-stamped to the second. ECI returns supersede the call when posted.
"When you can see the count, the count must be measured the way you would want to be measured yourself."
The tracker is the public mirror of our internal control room. Calls are made when the projected winning margin is > 1.5σ of the residual unmodelled variance for that constituency. Calls are signed by the desk partner on shift.
The points where the model re-projected the alliance total by more than �5 seats. Each point is a re-rake event triggered by a regional surprise.
Every estimate on this page passes the same standard as our published waves � margin disclosure, named researcher, signed re-statement when ECI returns supersede.
Constituency calls are issued only when the projected winning margin exceeds 1.5σ of unmodelled variance for that seat. State and alliance totals are re-projected on every constituency call via a fast incremental rake against the same five frames the National Pre-Poll uses.
If the count produces a result outside our 95% interval for any state, the desk halts and re-rakes from scratch before publishing the next state total. That halt is logged. Our 2024 GE halt log is published in full on the methodology page.
The tracker is the public mirror of our control room. The full method — sampling frame, weighting, calibration, monte-carlo, error log — is published in the next dossier.
