The Riksdag 2022–2026

How should the parties have voted — according to their own documents?

For every chamber vote in the Swedish Riksdag 2022–2026, one AI agent per party decides which side that party should have backed — based solely on the party's own election manifesto and party programme, plus a point-in-time snapshot of the country. Bloc documents such as the Tidö agreement and budget motions are deliberately excluded: each party is read against its own plan alone. The stance is then compared with how the party actually voted.

This is reconstruction, not prediction: the model's training data likely contains the real outcomes. Contamination is measured and reported openly. The project is politically unaffiliated.

51%of votes carrying an explicit commitment in the plan went against that commitmentn = 3325

Evidence tiers

  • 3939 explicit commitment
  • 5953 derived from principle
  • 10396 plan does not address this
JournalistWhere do parties break from the plan?

Every vote where the AI's document-faithful vote differs from the real one — sorted, searchable, with sources and embeddable graphics.

ResearcherMethods, data and reproducibility

Full methodology, contamination defence, open prompts, raw data and code.

CuriousSearch a topic you care about

Find the vote, see how your party voted — and what its own plan pointed to.

Evidence

Every claim can be followed in both directions

The quotes are not decoration — they are an index that reads both ways. 77,522 citations tie 4,715 lines of 23 party documents to 2,536 divisions.

From a vote to the document

Every AI decision has to quote the party’s own documents verbatim. The quote links to that exact line, highlighted where it stands.

Open a vote →
From the document to the votes

Open a party document and every cited line carries a margin note: which divisions leaned on it, and whether the party then voted with its own plan or against it.

Open a party document →

By party

How often did parties vote against their own plan?

The same thing, but across every decision the agent could read a position from the plan for, including where the plan only touches the question in principle. More decisions per party, weaker evidence per row.

KDKristdemokraterna77%1943 of 2535LLiberalerna75%1890 of 2535MModeraterna53%1342 of 2535SDSverigedemokraterna52%1199 of 2311CCenterpartiet49%955 of 1949SSocialdemokraterna45%985 of 2203VVänsterpartiet40%822 of 2057MPMiljöpartiet39%729 of 1886

Explore

Navigate by time

Each cell is a month: number of votes, colored by AI ↔ actual agreement (green = follows the plan). Click for that month's cases.

All cases (2539) →