The record.
The context.
Your judgement.
Stances brings together what politicians say and do, and the interests and relationships around them. It gives you a place to start looking, with a route back to the evidence.
Public records, brought together.
We collect parliamentary speeches, questions and votes, declared financial interests, staff interests, expenses and published ministerial meetings. Company records help put organisations and directorships in context.
Records come from UK Parliament, the Scottish Parliament, government publications, IPSA and Companies House. These are the publishers; Stances is an independent project.
We match records to people and organisations using published identifiers where available, alongside name matching. Names can be ambiguous, and a match can be wrong. A company connection describes a recorded relationship; it does not establish a shared agenda.
Social posts appear where they have been imported and attributed to an account. Their source and date matter: a post, a quotation and a parliamentary vote are different kinds of evidence. Social topic and messaging summaries are separate from the parliamentary stance and relationship scores.
Analysis stays linked to the underlying record.
Automated rules and language models group material by topic and identify expressions of support, opposition or a mixed position. A topic label says what something concerns. It does not, on its own, say which side the speaker takes.
Where available, position summaries include excerpts or reasoning so you can inspect the basis for a conclusion. Models can miss context, including irony, quotations and procedural language. Confidence describes the model's assessment; it is not a guarantee of accuracy.
How should I read votes and positions?
A vote records a choice on a particular motion, sometimes an amendment or a procedural question. An Aye is not automatically support for an entire policy. Speech–vote differences are prompts to read both records, with their dates and wording, before drawing a conclusion.
What do relationship signals mean?
These experimental indicators compare topics of parliamentary activity with donor sectors, recorded employment, staff interests and available position or voting evidence. They help surface records worth examining. A score is not a probability that money influenced a politician, or evidence of wrongdoing. Shared interests, party policy and constituency concerns can all explain an overlap.
Missing inputs and different reporting periods limit comparisons. A low or unavailable score cannot establish the absence of a relationship. Follow the component evidence and dates rather than using the score as a ranking of politicians' integrity.
What do organisation and messaging summaries mean?
Organisation profiles bring together the meetings, donations and directorships we have matched. The cross-source indicator describes representation across record types, not power or influence.
Messaging summaries describe topics and recurring frames in the posts analysed. Sampling, platform availability and the selected period shape the result; it is not a complete account of what a politician believes.
An expanding collection.
Coverage varies by person, source and period. Some records have not been imported, and some collected material has not been analysed. Blank results mean we have nothing to show for that selection; they do not mean nothing happened.
The coverage page shows the records currently held. A marked month contains at least one dated record, not necessarily every event in that month. Source limits, gaps and undated records are shown separately. Published records can also arrive late or be corrected.
Keep the source close.
Before quoting a finding, open the original record where linked, check its date and context, and distinguish the publisher's words from our analysis. Source links are not available for every legacy record.
If you find a bad match, a missing source or a misleading summary, report a correction with the profile link, the source and a brief explanation. GitHub reports may be public; include only information you are comfortable sharing.
Methodology overview · September 2026. Updated alongside changes to the public app.