How this works.
Trust in an election compass comes from being inspectable, not from claiming it’s neutral. Here’s the whole pipeline.
Where the questions come from
Questions are drawn from each party’s published 2026 manifesto, scoped to topics where parties actually disagree. The wording is LLM-assisted, with the goal of plain language, neutral framing, and no double negatives. The full question pool is browsable in the compass.
Where the party positions come from
Every position is read directly from the party’s published 2026 manifesto:
Each position is extracted from the party’s 2026 manifesto by an LLM and rated on a 5-point scale. When the manifesto doesn’t address a question directly, the position is marked as inferred from the party’s broader public platform.
Every entry carries a verbatim Swedish quote from the manifesto and a section reference, so the source can be checked directly on the site.
Each position is tagged high / medium / low confidence based on how directly the manifesto addresses the question. Low-confidence positions are downweighted in matching.
“A compass that asks for your trust should be able to show why it deserves it. Every step of the math, every quoted line of every manifesto.”
How matches are calculated
For each party, the match is the weighted similarity between your answers and the party’s positions. The weight wᵢ on each question is the product of three signals:
wᵢ = salienceᵢ · discriminationᵢ · confidenceᵢ
match(party) = Σᵢ wᵢ · (1 − |userᵢ − partyᵢ| / 4) / Σᵢ wᵢ- Your salience
- 1× by default, 2× when you mark a question as important to you.
- Question informativeness
- σ of the parties’ positions on the question, divided by the mean σ across the question pool. Questions where every party agrees count for almost nothing; questions that actually separate the parties count for more. Floored at 0.25×.
- Manifesto clarity
- High 1.0 / medium 0.7 / low 0.4 — parties with vague manifestos can’t ride high on confident matches.
The confidence band shrinks the more you answer (roughly 1/√n) and widens for parties whose positions were harder to read clearly.
How the dashboard aggregates polls
A daily run pulls numbers from publicly available pollsters (Novus, Demoskop, Verian/Sifo, Ipsos, SKOP, Sentio) via attribution-respecting sources. The 30-day average is sample-weighted (√n). Pollster house effects are shown rather than averaged away. Every number links back to its source poll.
Sources
- Election results
- Valmyndigheten (val.se). Open data, historical results down to precinct level.
- Demographics
- SCB pxweb API. Statistics Sweden’s official data API.
- Riksdag votes
- Riksdagens öppna data (data.riksdagen.se).
- Polls
- Aggregated daily from Wikipedia’s continuously updated polling page (CC BY-SA), with attribution.
- Manifesto coding
- Manifesto Project (MARPOR) and Chapel Hill Expert Survey (CHES) used as cross-references.
Photo credits
Leader portraits come from Wikimedia Commons. Each image is shown under the license listed; click a name to open the original file.
What this is — and what it isn’t
This is an independent civic project, not an official source. Please read this before you act on anything you see here.
- How positions are read
- Every party’s positions are AI-interpreted from its published 2026 manifesto, with a verbatim Swedish quote and a page reference. The quote is the authoritative source — the summary may contain interpretation errors. When it matters for your vote, read the quote.
- Provided as is
- Everything on this site is provided as is, without warranty of accuracy, completeness, or currentness. You use the compass at your own discretion. We accept no liability for decisions made on the basis of what is shown here.
- Not affiliated with
- polara is not affiliated with any political party, the Riksdag, Valmyndigheten, or any government body. This is an independent civic project, not connected to any party’s campaign organisation.
Ready to start?
Start the compass