| Party | Economic | Social | eSafety |
|---|
Primary votes are the AEC's final 2025 results, held at candidate level rather than pooled by party. Every division not affected by a 2026 redistribution uses the AEC's final 2025 two-candidate-preferred count as both its TCP baseline and its final pair, so the model reproduces all 150 published results exactly when nothing has been moved.
South Australia, the ACT and Tasmania use Antony Green's final notional estimates on the 2026 boundaries. Notional margins use his final 2026 redistribution calculations, and seat hover notes show the old-to-new margin change for those three.
Tasmania's final boundaries were announced on 15 July 2026 and all five names were kept on 10 August 2026. South Australia and the ACT are also final. Grey is displayed by its incoming name, O’Donoghue.
Queensland's federal redistribution formally commenced on 16 September 2026, after being deferred in August 2025. The AEC's determination found 20 of Queensland's 30 divisions outside the permitted enrolment range, so most Queensland boundaries will change. The notice of determination is due to be published on 22 September 2027.
New boundaries apply only to an election whose writs are issued after that date. Whether Queensland is contested on these 2025 boundaries or on new ones therefore depends on when the election is called. All 30 Queensland divisions here remain on the 2025 boundaries, and their margins will change if the redistribution takes effect first.
Where a redistribution moves 2025 votes for a candidate into a seat that candidate did not contest, those votes stay with that candidate as a separate count entry instead of being merged into another candidate or reassigned to a party. The model makes no assumption about how that candidate would have performed in the new seat; it simply keeps the raw votes where the voters cast them.
| Division | Ghost candidate | Share |
|---|---|---|
| Lyons | Andrew Wilkie (Glenorchy area) | 19.2% |
| Clark | Peter George (Franklin area) | 11.15% |
| Clark | Brendan Blomeley (Franklin area) | 1.8% |
| Bean, Canberra | Jessie Price / Claire Miles cross-over | — |
| Bean | Teresa McTaggart (Animal Justice), Mary-Jane Liddicoat (HEART) | — |
| Bass, Franklin | Angela Offord (Lyons area) | — |
| Franklin | Carlo Di Falco (Shooters, Fishers & Farmers), Michael Phibbs (Citizens) — Lyons area | — |
| Barker, Boothby | Centre Alliance (from Mayo) | — |
The Clark Wilkie/George split (26.05 / 11.15 of Green's 37.2 Independent total) and the ACT split are estimated from Green's sub-area figures and AEC enrolment for the moved areas, not derived from booth files.
Every Independent, and every candidate from a party without its own model category, is a separate count entry. Candidates are therefore excluded one at a time rather than pooled — Calwell's Moore, Youhana and Moslih are three separate exclusions, not one.
Votes are tracked by the candidate they started with, so an excluded candidate's pile moves by the preferences of the voters who cast each parcel, not by the flows of whichever candidate happens to be holding it.
When three or more candidates remain, survivors are first grouped by the party the chart sees them as — two Coalition candidates, or two Independents, count as one chart party. Each parcel is split between those parties using every pair’s chart value: with two parties left this is exactly the chart percentage, and with three or more the pairwise ratios approximate the chart (exactly, when the chart values are mutually consistent). A party’s share is then split among its own candidates in proportion to their votes at that point in the count, so a party with two candidates never receives double the preferences.
Final-two candidate identity is determined by the preference-flow chart through a full elimination count, with modelled 4CP and 3CP shown for every seat. The TCP between the two finalists is the direct pairwise transfer of every other candidate, using the chart. There are no anchors or bonuses: while the model is untouched each seat shows its published 2025 TCP (the real-count pin, section 5), and once anything moves the chart’s own figure is shown.
On an exact tie at an exclusion, the candidate lower on first preferences is excluded. This simplifies the AEC rule, which looks back to the last count at which the two differed. If they are also tied on first preferences, the choice is alphabetical. If every other candidate has been swung to zero, the division is shown as uncontested: the remaining candidate on 100% against the strongest 2025 challenger on 0%.
In six divisions — Blaxland, Calwell, Hunter, Lyons, Maranoa and Newcastle — the preference table on its own does not reproduce the actual or notional final two. While the model is untouched, those divisions are pinned to the published final two. Every division’s TCP, and the national panels, are pinned to the published figures in the same way.
The pin is a small exclusion-order calibration applied only while every Swing Lab control sits at its default and no preference flow differs from its default. It is shown in the seat hover, and the Swing Lab reports whether the pin is on or released. Moving any control, or changing any flow, releases the pin for the whole model; from that point the preference table’s own count decides the final two in every division, including those six. The same pin holds the Senate’s New South Wales and Victorian results (see section 8). Typing a flow back to its default value, or using Reset, turns the pin back on.
This is deliberate. No set of nationally consistent, evidence-based preference flows reproduces these divisions. The values that would — Family First at 100% to the Coalition over an Independent in Blaxland, right-minor parties at 0% to Labor over One Nation in Maranoa — are outside anything observed in a real count. The model shows the published result rather than inventing flows to manufacture it.
All model defaults are multiples of five; custom edits move in 2.5-point steps. Two-way transfers are the only mechanism — there are no bespoke per-seat flows anywhere in the model.
| Contest | Donor | Value |
|---|---|---|
| Labor v Independent (% to Labor) | Coalition | 30 (29.3 over six counts; Watson’s Liberals split 52.5% to Labor) |
| One Nation | 30 (28.0; replaced a 35-to-20 modelled gradient) | |
| Family First | 30 (28.8) | |
| Trumpet | 35 (37.2) | |
| Libertarian | 30 (30.8) | |
| Another Independent | 45 (47.3) | |
| Coalition v Independent (% to Coalition) | Labor | 25 (22.6) |
| Greens | 10 (12.4) | |
| Family First | 65 (62.8) | |
| Trumpet | 50 (52.2) | |
| Legalise Cannabis | 40 (39.4) | |
| Libertarian | 75 (75.4) | |
| Another Independent | 30 (30.6) | |
| Pooled Shooters / People First bloc | to Coalition over IND | 45 |
| to Labor over IND | 45 | |
| to Labor over ONP | 35 |
The pooled right-minor bloc is looser than the Family First / Trumpet average the model would otherwise derive for it, which is what the 2025 counts show.
A measured row replaces a modelled one only where 2025 provides at least two divisions, or is the only count of that type that exists. Single small samples are left on the national interpolation rather than allowed to override it on a few thousand votes:
| Left out | Sample |
|---|---|
| Legalise Cannabis at Calwell | 2,843 votes |
| Socialist Alliance at Fremantle | 970 votes |
| Fusion at Cowper | 255 votes |
Geography variation is used only where the evidence shows a changing right-bloc relationship, and only in two chart cells. Coalition-to-One Nation in Labor v One Nation contests runs from loosest in the cities to tightest in the country: 60 in Inner Metro, 65 Suburban, 70 Outer Metro, 75 Provincial and Tree-change, 80 Rural (Enfield 53, SA’s 26 contests about 68 overall, Secret Harbour about 70, Hammond 73.5, Hunter 82.9). Tree-change has no Labor v One Nation count of its own and follows Provincial, because in 2025 its right-bloc voters preferenced like country voters. One Nation-to-Coalition against an Independent runs the other way — 75 in Inner Metro and Suburban, 65 in Outer Metro and Tree-change, 60 Provincial, 55 Rural — because 2025 One Nation voters in teal seats (Kooyong 77%, Mackellar 73%) backed the Liberal far more strongly than in rural independent contests (Farrer 53%, Calare 56%). One Nation-to-Coalition in Labor v Coalition contests is flat at 75: across 112 divisions in 2025 every geography class sat between 71 and 78, with no consistent city-to-country order (74.3 overall; SA 2026 69). Every other row is flat.
There is no CA v KAP column: Centre Alliance votes exist only in Mayo (plus ghost Mayo-area votes in Barker and Boothby) and Katter candidates only in Herbert, Kennedy and Leichhardt, so that contest cannot arise in any division.
The three national TCP panels are calculated by vote-weighting and aggregating the 150 divisional pair counts, matching the AEC's own national TPP aggregation method. SA, ACT and Tasmanian divisions use their post-redistribution notional baselines.
Once anything moves, every panel is the model’s own aggregate with no correction. For Labor–Coalition that starts about 0.1 points below the published figure, so the first touch of any control shifts it slightly.
The state cards show seats and primary votes per state. The primary column includes an Other residual so every state sums to 100%.
The chamber is 76 seats: the 40 contested at the next election plus the continuing senators. Because the next election falls when the 2022 cohort's terms expire, the continuing senators are exactly those elected in 2025.
The projection uses a simplified group-level STV count: a Droop quota of 100/(seats+1), full quotas banked first, then the lowest group is excluded one at a time. A party’s leftover above its full quotas stays with it as its next candidate’s vote — which is what happens in a real count, because above-the-line votes follow the party’s own ticket.
Where excluded votes go. Each Senate group is read as its party on the Preference Flows chart, and the split among the groups still in the count follows that chart exactly as in the House, averaged over the state’s own mix of geography classes — so editing a flow moves the Senate too. Senate-only minor parties are placed by where they sit on the left-right spectrum: Socialists, Fusion, Indigenous-Aboriginal and Australia’s Voice as Minor left; Christians, Shooters, People First, HEART, Citizens, Great Australian and Katter as Minor right; Pocock, JLN, Lambie and ungrouped candidates as Independent; Sustainable Australia and the Democrats as Others.
Exhaustion. Parties sit on a five-step line: left (Greens, Legalise Cannabis, Animal Justice, Minor left) · Labor · centre (Independents, Others) · Coalition · right (One Nation, UAP, Family First, Libertarian, Minor right). The share of an excluded party’s votes that exhausts depends on how many steps away the closest party still in the count is:
| Steps to the closest party still in | 0 | 1 | 2 | 3 | 4 |
|---|---|---|---|---|---|
| Left-wing and Labor votes | 5% | 15% | 30% | 50% | 70% |
| Centre votes | 10% | 20% | 40% | – | – |
| Coalition and right-wing votes | 10% | 20% | 40% | 60% | 80% |
Early in a count almost every voter’s next choices are still in, so little exhausts; late in a count, when only parties a voter dislikes remain, most of it does. Left-wing voters number each other’s parties more reliably than right-wing voters do, so their row is lower at close range. These are judgement values in 5-point steps, not measured from Senate counts.
Real-count pin. At default settings this count reproduces the 2025 result in six states, but gives New South Wales’ last seat to Labor instead of One Nation and Victoria’s to One Nation instead of Labor. No single consistent set of flows fixes both — the 2025 Senate flows differed by state, most likely because of state-specific how-to-vote cards. So, exactly as in the House, each state shows its actual 2025 result until any Swing Lab control or preference flow is changed. Once the pin is released, New South Wales needs roughly a 1.3-point One Nation gain, and Victoria roughly a 0.5-point One Nation drop, to return to the 2025 outcome.
The Senate vote responds to the House sliders through each state’s aggregate House shares. Every point of House independent movement is counted once. Pocock and JLN have no House ticket, so they take half the movement in their state’s independent vote excluding divisions held by a sitting Independent MP, bounded to ±40% of their own Senate vote; the rest of that movement, and all independent movement in the other states, becomes an ungrouped Senate group (“Ungrouped IND”) rather than being split across the state’s minor parties.
There are no calibration bonuses.
All controls move in locked 2.5-point increments and are expressed in points of the division total. Labor is the balancer: in each seat Labor moves by minus the swings actually applied to the other candidates who stood there, so a party that did not contest a seat takes nothing from Labor in that seat, and a party asked to lose more votes than it holds cannot hand Labor the difference.
The Apply poll average button moves every seat’s 2025 primary vote to a weighted average of the latest federal polls, before any slider is applied. Only the most recent poll from each company is used. Each poll is rescaled to 100% and weighted by pollster tier (Newspoll and YouGov 1.0; RedBridge and Resolve 0.8; DemosAU and Essential 0.6; Roy Morgan, Spectre and any other firm 0.4) times recency, which halves every 14 days counted back from the newest poll. The poll list is refreshed from Wikipedia’s opinion polling page every three hours (Sydney time); a firm whose latest poll is more than 90 days older than the newest poll is left out. Polls report Labor, L/NP, Greens, One Nation and Others; Others includes Independents and every minor party.
Within a group the change is shared among its candidates by their vote (Liberal and National candidates in the same seat, or several Independents). No vote goes below zero, each seat is rescaled to 100%, and the swing is solved so the national primaries match the average exactly. Using the poll average releases the real-count pin.
Starts from the AEC classes, then moves selected seats into a Suburban class and selected lifestyle seats into Tree-change. Tree-change marks regional, peri-urban and bushland seats with a substantial metro-origin population and a pro-climate-action bloc inside the electorate, which makes them less conservative than typical regional or outer-suburban seats even where the seat as a whole is not wealthy. The set is Berowra, Casey, Corangamite, Eden-Monaro, Flinders, Forrest, Franklin, Gilmore, Indi, Lyons, Macquarie, Mayo, Page, Richmond and Ryan. Moncrieff, McPherson and Solomon are treated as Suburban — Darwin’s urban form has little in common with the Inner Metro seats such as Melbourne, Sydney, Grayndler and Brisbane that the AEC class would otherwise put it with.
Full ED/EA seats receive 100% of the relevant economic slider; the 25 ED1/2 seats and 25 EA1/2 seats receive 50%. All other seats receive no economic-profile swing.
A 2023 Voice-referendum proxy using the No share: Socially progressive (below 50%), Socially liberal (50–57.5%), Socially moderate (57.5–65%) and Socially conservative (over 65%), with current-boundary overrides where a redistribution materially changes the character of a seat. Current-boundary judgement overrides apply to Bennelong, Hotham, Boothby, Bullwinkel, Bass (socially moderate) and Franklin (socially liberal).
Tags are mixed rather than mutually exclusive: a seat with one tag receives 100% of that slider, two tags 50% each, three tags one-third each, four tags 25% each. Small business owner is the fifth option and uses the same dilution rule as Blue collar, White collar, Pink collar and High unemployment; its working 70-seat set runs from O’Connor through Fadden inclusive in the estimated owner-manager ranking.
ABS 2021 CED tenure shares with a sequential 5.0 percentage-point mixing rule. Bullwinkel is estimated from predecessor electorates.
Things the model does not do well, stated plainly.
Calwell needs a pin under the current engine. Five Independents stood there in 2025. Now that same-party candidates share one party’s preferences, split by their votes, Carly Moore finishes about 1.3 points short of the Liberal and the model gives Labor v Liberal, where the real count was Labor v Moore. Earlier versions gave each Independent a full share of independent preferences, which is what carried her through. That double counting was wrong, so Calwell is absorbed by the pin instead.
The last Senate seats in New South Wales and Victoria are pinned, not modelled. At default settings the chart-based count swaps them (Labor instead of One Nation in NSW, One Nation instead of Labor in Victoria), so both show their actual 2025 result only until the Swing Lab or a preference flow is changed. Measuring the real 2025 Senate flows from the AEC distribution-of-preferences files would be the fix.
Senate minor parties are close to frozen. UAP and Legalise Cannabis are driven by House primaries of about 1.9% and 1.2%, which no slider can move far, so their Senate votes stay near their 2025 values regardless of input.
Ghost Independent splits are estimates. The Clark and ACT splits come from sub-area figures and enrolment, not booth files, and could be materially wrong.
Queensland is on stale boundaries. 20 of its 30 divisions are outside the enrolment range and will change if the redistribution takes effect before the election is called.
Swing overlays overlap. Geography, demographic, social, workforce and tenure are each exhaustive, so moving every slider in any one family is equivalent to a uniform national swing. The 23 sliders span 19 independent directions, and correlated pairs — affluent / white collar / socially progressive, for example — stack on largely the same seats.
Swings are in points of the division total, applied across the model's geography buckets. Labor is the balancer for the party/geography controls — in each seat Labor moves by minus the swings actually applied to the other candidates who stood there, so a party that did not contest a seat takes nothing from Labor in that seat. The Coalition umbrella control moves the combined Coalition vote by the slider amount, shared across whichever Coalition candidates (Liberal, LNP, Nationals or CLP) actually stood in a seat; Liberal/LNP/Nationals controls remain available as additional party-specific adjustments. A party swing cannot conjure a candidate into a seat it never contested (the Other slider is the one exception, described below). Mixed demographic, economic-profile, social-profile, workforce/labour-market and housing-tenure controls are separate seat-bucket overlays; the social tiers use the 2023 Voice No share as a coarse proxy, with current-boundary adjustments. Candidate-effect controls are separate seat-level overlays: sophomore incumbency, confirmed retirement, and an optional reduction to 2025 Independent candidates who are not sitting Independent MPs. The pooled Other vote also has a global all-seats slider, defaulting to 0 points and divided equally among Other candidates present. All Swing Lab controls move in locked 2.5-point steps. Published 2025/redistribution TCPs remain empirical baselines. All model preference defaults are multiples of 5; only editor changes use 2.5-point steps. Geography is used only for two right-bloc flows (see the Methodology tab); every other flow is flat.