Track Record
Every forecast, scored against the result
A forecasting site earns trust in only one way: publish before the event, score after it, and show the misses with the hits. This page is where that happens.
Status
No 98th Percentile forecast has yet resolved. The first test is the Victorian state election on 28 November 2026. The first-preference estimates published during the campaign, and the victory forecasts to follow once the how-to-vote cards are registered, will all be scored here in December 2026: every published figure against the certified result, using the scoring rules set out below before the election, with nothing revised after the fact and nothing quietly dropped.
Until then this page makes no accuracy claims.
How forecasts will be scored
Point estimates are compared against certified results as published by the Victorian Electoral Commission. Interval coverage is reported against the stated level: a 90 per cent plausible range should contain about nine results in ten, and the page will report how many it actually contained. Win probabilities, once published, are scored with proper scoring rules and binned calibration, against a simple baseline fixed before the result is known. Each comparison cites the run identifier of the forecast as it was published at the time, never a revised version.
Which release counts
Every release is scored. None is dropped. Forecasts update as new polls arrive, so an election produces many releases before it resolves. For each election the headline release is named in advance, before any result is known.
For Victoria 2026 the headline is the final release before polling day. The first release of win probabilities, published after the how-to-vote cards are registered, is reported as a fixed earlier point. Every other release is scored and published beside them.
Standing forecasts, such as the cash rate or the CPI, never finish. Each will be scored every time it resolves, and the result will enter this page as it lands.
Release archive
| Release | Election | Published | Scored |
|---|---|---|---|
| First-preference estimates, all 88 electorates Updated as polls pass review |
Victoria 2026 | 27 October 2026 | December 2026 |
| First win probabilities and seat totals Named early scoring point |
Victoria 2026 | After 9 November | December 2026 |
| Final pre-election release Headline release |
Victoria 2026 | By 27 November | December 2026 |
Back-tests are not a track record
One out-of-sample test exists and is reported for what it is: an earlier, simpler version of the electorate-level method, applied to the 2026 South Australian election, reproduced each electorate’s One Nation vote to within 1.7 percentage points on average. That is a back-test of a predecessor method on one election, labelled as such on the Methodology page. It appears here so that it is never mistaken for a scored forecast.
Use of AI
I used large language models, Anthropic’s Claude Opus 5.5 and Claude Fable 5, to test this work for narrative flow, logical consistency, bias and errors. They also assisted with drafting and with the site’s code. Responsibility for the content is mine alone.
John Ward (The 98th Percentile)