BBC PROJECTED NATIONAL SHARE · 8 MAY 2026 · pre-registered hash on GitHub before voting opened

Most accurate national vote-share prediction in the UK polling field. We did it with a few hundred quid of compute.

On the BBC's official Projected National Share for the 7 May 2026 English local elections, KPM-1 had the lowest mean absolute error of any model in the field. We pre-registered the predictions with a SHA-256 hash on GitHub before any ballot was cast. The hash receipt does not move.

2.96pp
KPM-1
3.20pp
YouGov
3.20pp
Opinium
3.50pp
10-poll consensus
5.00pp
Electoral Calculus

Mean absolute error vs BBC PNS, lower is better. KPM-1 was the closest pollster on Labour, Liberal Democrat, and Green vote shares; close second on Reform UK. We were also beaten on council-control accuracy by a specialist heuristic — the post-mortem covers both findings honestly.

Read the post-mortem → View the predictions How the engine works

Hash: 1fd2be14dc6e0148… · GitHub: kpm1-election-projections · Validation paper: DOI 10.5281/zenodo.19361059 · Commercial: jason@kronaxis.co.uk

20
Countries
187
Fields per persona
8
Personality dimensions
BSL
Source available

Panel Studio

Consumer panels on demand

Build and query synthetic consumer panels. Define a demographic, generate personality profiled respondents, and run multiple turn conversations. Every response includes a causal reasoning trace explaining which personality dimensions drove the decision.

Census Weighted Panels
Each persona's demographics are sampled from real census data across 20 countries. Panels match actual population distributions for age, ethnicity, education, and regional deprivation.
Causal Reasoning Traces
Every response includes a structured trace: which personality dimensions were active, which memories influenced the decision, the emotional state, and the financial context. Exportable as JSONL.
Self Hosted, No Cloud
Runs entirely on your infrastructure. Local LLM inference. No API keys to external services, no data leaving your network. Docker Compose with zero configuration on first boot.

How it works

From deployment to data

01
Define audience
Country, demographics, geography, personality distribution
02
Build panel
Census weighted personas with DYNAMICS-8 profiles and life histories
03
Stimulus testing
Submit your message, product, or concept. Every persona responds in character
04
Results and data
Quantitative analysis plus DYNAMICS-tagged reasoning traces for AI training

Open specification

DYNAMICS-8: the personality engine

Eight dimensions purpose built for behavioural simulation. Each is a continuous float from 0.0 to 1.0 with four granular facets, giving 32 behavioural parameters per persona. Published under Creative Commons Attribution 4.0.

D
Discipline
Y
Yielding
N
Novelty
A
Acuity
M
Mercuriality
I
Impulsivity
C
Candour
S
Sociability
Read the full specification Take the personality quiz

Use cases

What people build with Panel Studio

Market research
Run a 500 person synthetic panel against any creative, pricing scenario, or product concept in hours. Understand why different segments respond differently.
AI training data
Causal reasoning traces tagged by personality, exported as JSONL. Each record documents why a specific type of person chose a specific response.
Academic research
Census weighted panels across 20 countries for policy simulation, election prediction, and social science research across cultures.
Brand strategy
Test messaging across personality segments before launch. See which DYNAMICS dimensions predict positive response and which drive rejection.
Product testing
Show your product to synthetic consumers with realistic life contexts. Understand purchase intent and price sensitivity segmented by personality type.
Pricing research
Conjoint analysis with personas that reason about affordability through their own financial context, risk tolerance, and spending habits.

Platform

More than panels

Panel Studio is the source available core. Kronaxis also provides government capabilities, open datasets, and free tools for personality science.

Open datasets
1,000 sample personas on HuggingFace (500 UK, 500 US) with 187 fields each. DYNAMICS-8 specification published under CC BY 4.0. Python and Go reference implementations. Full panels across 20 countries through Panel Studio.

Browse releases →

Government and Defence
Autonomous digital operations, security evaluation, and intelligence capabilities for UK government and defence. Sovereign infrastructure, UK patent protection, no foreign cloud dependencies.

Learn more →

Free tools
Discover your DYNAMICS-8 personality profile with our quiz. Experiment in the interactive playground. Take the certification course. Build with the developer sandbox. All free, no account required.

Take the quiz →

Quick start

Running in five minutes

Clone the repo, start Docker, and you have a working panel with 500 personas loaded. No registration required for non commercial use.

# Clone and start
git clone https://github.com/kronaxis/kronaxis-panel-studio.git
cd kronaxis-panel-studio
cp .env.example .env
docker compose up -d

# Open Panel Studio
open http://localhost:8090

# Optional: add your API key for premium features
# Get one free at kronaxis.co.uk/register
echo "KRONAXIS_API_KEY=kx_your_key_here" >> .env

Built on the level

Protected by UK patent applications (4 filed)
Self hosted: your data stays on your infrastructure
UK sovereign technology, designed and built in England

Start building panels today

Free for research and personal projects. Get an API key to unlock premium features, or clone the repo and run it locally with no registration.

Get a Free API Key View on GitHub

Contact

Get in touch

For enterprise licensing, research collaborations, dataset access, or technical questions.

Contact options

We respond to all enquiries within two business days.

  • General enquiries
    contact@kronaxis.co.uk
  • Registered office
    England & Wales, United Kingdom

Your details are used only to respond to this enquiry. We do not share contact information with external parties.