Show HN: 65k AI voters predict UK local elections with 75% accuracy

6 hours ago

65,000 synthetic UK voters were created from Census 2021 data, each with demographics, a DYNAMICS-8 psychological profile, and political history to predict the 7 May 2026 local elections.
The model was validated against 10 real council by-elections in March 2026, improving accuracy from 1 to 6 correct winners out of 8 after nine iterations and calibration.
Calibration revealed systematic errors: over-predicting Reform UK by ~10% and under-predicting Liberal Democrats by ~7%, applied to predictions.
Predictions for the first 20 councils (all currently Labour-held) show Reform UK leading with 39.6% vote share, winning 18 councils, while Conservatives collapse to 7.7%.
Key specific predictions: Doncaster and Sunderland see Reform UK above 50%, Bristol goes Green, Manchester holds for Labour narrowly, and Birmingham, Coventry, and Sheffield are Reform blowouts.
Findings indicate Reform UK's support is durable, driven by personality traits like low Novelty and high Discipline; Greens attract progressive voters; Lib Dems are strong in affluent areas.
Methodology involves ward-level matching, multi-layer correction (turnout, incumbency, protest dynamics), and inference via Kronaxis Imprint, a 27-billion parameter language model.
Full predictions for 136 councils will be published on 1 May 2026, with an accuracy report after results and open access to methodology and DYNAMICS-8 model.

Hasty Briefsbeta