It's election eve, and the exclusive Herald Election Forecast has Winston Peters as the king- or queen-maker - as it has since we launched the model a month ago.
The model has been updated for the final time following the release last night of the final poll of the campaign.
The Newshub Reid Research poll, conducted between September 13 and 20, put National on 45.8 per cent with Labour on 37.3 per cent. The Green Party had moved from below the 5 per cent threshold to 7.1 per cent. New Zealand First was also on 7.1 per cent with the balance of power.
The Herald's Election Forecast crunches data from every major poll conducted in the past year and the results of every election since 1999. Check out the latest update in full, and find out how it was built here.
We can't guarantee it's going to predict exactly what will happen tomorrow, but statisticians believe that collecting more data, over a longer period, is more likely to create a true picture of where New Zealand's voting intentions lie.
The data is fed into the model, which simulates thousands of outcomes. The possible outcomes give an estimate of most likely result along with lower and upper estimates.
Our final forecast puts National on 56 seats in parliament, with a Labour-Green coalition at 54 seats.
NZ First is now projected to get 7.5 per cent of the party vote and nine seats.
The estimates are median estimates. The lower and upper estimates are shown in the visualisation and give better estimates of uncertainty surrounding the projections.
To form a government, a party or coalition needs at least 61 out of 120 seats, or to reach a confidence-and-supply agreement.
Individual polls are showing more volatility than normal due to leadership changes.
But at this late stage, the Herald Election Forecast is largely the same as it was at the beginning - it is highly unlikely that Labour or National will form a government without NZ First support.
The model also includes predictions for candidates by every electorate. These are based on previous election results and take into account any polling that has taken place for a particular electorate.
There is different methodology for Epsom, which is unlike most other electorates because of tactical voting.
The visualisation also includes candidate and party vote predictions for Maori electorates. However, this is perhaps the hardest part of forecasting because of tactical voting and lack of consistent polling at electorate level.