Methodology

The Race for No.10 reports an existing published rolling average of British Polling Council members’ voting-intention polls, rather than computing an average of our own. That is deliberate: you can check our headline number against its source in one click, and we cannot quietly put a thumb on the scale by choosing which polls to include.

The individual polls behind the average are listed on the front page with their pollster, fieldwork dates and a link to the pollster’s own write-up. Not every poll in the average appears: some pollsters release to clients and aggregators without publishing a write-up of their own, and we would rather list fewer polls than cite a source that is not where the figure came from. Where a pollster did not publish a figure for a party, we print an em dash rather than a zero or an estimate.

What this cannot tell you: vote share is not seats. Under first-past-the-post a party on 12% nationally may hold five seats while a party on 21% holds over a hundred. We do not publish a seat projection, because the available models diverge sharply and get quoted as fact. Movements of a point or two between polls are usually noise, not news.

Each masthead carries a fixed position on a −10 (left) to +10 (right) scale. The rating describes the outlet, not the individual article, and it does not change from story to story. Positions draw on Ofcom’s news-consumption research, the Reuters Institute Digital News Report’s left–right placement of news audiences, and published perception surveys. Where those sources disagree, we take the midpoint.

We rate the outlet rather than the article on purpose. The news table is filled from live RSS feeds with no human in the loop, so per-article scoring would mean having a model judge the slant of every headline: precisely the point at which a site promising neutrality becomes the story.

What this cannot tell you: a right-leaning paper runs stories that damage the right, and a left-leaning one does the reverse. A rating is a prior, not a verdict on the piece in front of you. Broadcasters regulated for due impartiality sit near the centre by obligation, which is not the same as being free of editorial judgement. Reasonable people place several of these mastheads a point or two from where we have.

Two outlets below are rated but do not appear in the news table: The Times and The Spectator publish no open RSS feed. We list their ratings for reference rather than quietly dropping them, so you can see the full picture of who we do and do not carry.

Party positions use the same −10 to +10 axis, as a composite of economic and social policy. These are editorial judgements informed by published academic classification, each party’s own stated policy, and where commentators across the spectrum place them. They are not measurements, and we would not defend any single figure to the point.

Where a party disputes how it is characterised, the dispute is printed alongside the characterisation rather than resolved by us. Restore Britain is the clearest current case: political scientists classify it as far-right and ethno-nationalist, and the party rejects that description as a smear campaign. Both statements appear on its page.

What this cannot tell you: a single axis flattens real disagreement. A party can be economically interventionist and socially restrictive at once, and the number will hide that. Read the position note, not just the dot.

We publish six of these. Three measure sustained state-level pressure on the United Kingdom: Russia, currently 55/100, alongside Iran and China. Three measure the opposite: what the UK can currently rely on from the United States, NATO, and France and the EU. None of them measures the likelihood of open war, which remains low.

The three threat assessments answer the same six questions, with the same weights applied to all of them. That was not always true. Each state used to be scored on its own set of factors, which meant the composites could not honestly be compared: a fifth of Russia’s score came from a question about Russian capacity being absorbed in Ukraine, a factor no other state faced, and the result was that Russia scored below China. That was an artefact of which questions were asked, not a judgement about which state presses hardest on Britain.

Each factor is still scored 0–100 on its own evidence, and each shows the reporting behind it. Where a state has no evidence against one of the six, the factor still appears, scores low, and says the evidence is absent, dropping the inconvenient question is what broke the old model. Where the UK government has made a formal designation. Russia and Iran sit on the enhanced tier of the Foreign Influence Registration Scheme and China does not. We print it and score against it rather than around it. Where the state rejects the characterisation, the rejection is printed beside our assessment.

The three alliance assessments still carry their own factors and are not yet on a shared set. They are comparable with each other only loosely, and that is the next piece of work rather than something to gloss over here.

Threats and alliances are deliberately kept on separate scales. They share a method but not a meaning, and one rail carrying both would invite the reading that a strong alliance cancels out a hostile state.

The six questions and the weights every threat assessment carries: