12,224 Voters Told Us Why They're Switching. Here's What They Said.

When 435 Labour supporters tell you they are holding their nose and voting Lib Dem, that is not a number. It is a story about a party losing its base from the inside.

We ran 12,224 synthetic voters through a two-question protocol for all 159 English councils (136 all-out + 23 thirds elections) going to the polls on 7 May. First we asked who they liked. Then we asked who they would actually vote for. The gap between those two answers is where British politics lives right now.

The Headlines

But the headlines are not the point. The reasoning is.

What 12,224 Voters Told Us

Every synthetic voter was asked two questions. First: which party do you genuinely prefer? Second: who will you actually vote for on 7 May, and why? The gap between sincere preference and tactical vote, captured in full-text reasoning, is the dataset no other forecaster produces.

Council tax dominates the reasoning. 63% of voters mentioned it. Brexit and Reform were cited by 65%. Cost of living by 23%. Housing by 25%.

But the real insight is in the tactical switching. 17% of voters chose a different party from their sincere preference. These are the flows that reshape the map.

The Biggest Tactical Flows

In Their Own Words

These are not aggregated statistics. These are individual reasoning traces from individual synthetic voters, each grounded in census-weighted demographics, a DYNAMICS-8 personality profile, and local electoral context.

Every forecaster can give you numbers. Electoral Calculus, Fisher, Curtice: they all produce vote share projections. Some are very good. None of them can tell you why. None of them capture the internal deliberation: the moment a Labour voter decides they cannot waste their vote, or a Reform supporter realises the Conservative candidate has a better chance locally.

We capture all of it. 12,224 individual decision stories. Every sincere preference, every tactical calculation, every piece of reasoning.

How Synthetic Personas Reason About Their Vote

Each of the 12,224 reasoning traces captures how a synthetic persona articulates its voting decision. The themes below reflect what the panel generates when asked to explain its choices, not a survey of real voters.

A note on what this table does and does not show. Socially sensitive topics like immigration are likely underrepresented in LLM-generated reasoning, even when they drive vote choice. The 65% citing "Brexit / Reform" almost certainly encodes immigration sentiment expressed through party identification rather than naming the issue directly. Real voter surveys consistently rank immigration in the top five issues. What this table reveals is how synthetic personas frame their reasoning in local terms: council tax, housing, cost of living. The individual traces and tactical switching patterns below are where the real insight lives.

Council Control

The V3 projections now cover all 159 councils: the 136 all-out elections plus 23 councils electing by thirds. The thirds councils include major cities such as Bristol, Liverpool, Leicester, Derby, and Doncaster.

Reform UK is projected to take control of 77 councils from a standing start. Labour drops from 92 to 17, losing control in every region outside inner London. The Liberal Democrats nearly double their council count. The Conservatives lose a net 4 councils, holding on far better than Labour. The Greens pick up 3 councils, concentrated in areas with strong activist bases.

What Changed from V2

Our V2 projections covered the 136 all-out councils. V3 adds 23 thirds elections and replaces the single-question protocol with a two-question approach that captures both sincere preference and actual voting intention.

The direction is the same; the magnitude increases slightly for Reform and the Conservatives, decreases for Labour and the Lib Dems. The addition of thirds councils (which tend to be urban, Labour-held) amplifies Reform's council count because several of these seats are in precisely the post-industrial northern and Midlands cities where Reform is strongest.

Pre-registration

The full prediction dataset was hashed before publication. The SHA-256 hash of the complete 159-council prediction file is:

sha256:663c022b425b5f38c423c52a0dfaf0ff98d69b4dcab719c38ba4f38868368eb0

This hash was generated from the JSON file containing all council-level vote shares, seat projections, council control calls, and ward-level disaggregation. After 7 May, anyone can verify that the predictions published here match the pre-registered file by comparing hashes. We cannot change the numbers after seeing results.

The full prediction file, verification script, and reasoning trace data are available on GitHub: github.com/Kronaxis/kpm1-election-projections

What Happens Next

1 May: Full rerun with the latest national polling data. Updated predictions will be published and hashed, capturing any late campaign shifts.

6 May: Final pre-election update with ward-level disaggregation for the 20 earliest-declaring councils.

7 May, from 11pm: Live results tracker. We have identified the 20 councils that declare first and 8 bellwether councils with specific signals to watch. Real results will be compared against our projections in real time as they come in.

8 May onwards: Complete accuracy report. Every correct call and every wrong call, with analysis of where the model succeeded and where it failed. No cherrypicking. If we hit 65%+ accuracy on council control and national vote shares within 4 points per party, the experiment has worked. If we miss, we will explain precisely why.

Methodology: How We Project the 2026 Local Elections: Every Step Explained

Previous projections: V2 Projections (136 councils) | V1 Projections (20 councils)

GitHub: Full data, code, and pre-registration hash

DYNAMICS-8 framework: Zenodo preprint (CC BY 4.0)

Interactive results: Election Results Browser (national, regional, council, and ward level with reasoning traces)