VoxSim · Technical Whitepaper
Stop Polling.
Start Wargaming.
Directional fault-line maps of the electorate — in minutes, not weeks.
We tell you where to look, not what you will find. A strategic wargaming engine for decision-makers under time pressure. EU AI Act Annex III ready.
- Backtested
- Calibrated against 130+ historical political events
- Granular
- 312 demographic cells across 12 distinct political clusters
- Transparent
- Known systematic biases explicitly documented and quantitatively offset
The Problem
The "Undecided Voter" Illusion
Traditional polling consistently fails at the margins that decide elections. By pressuring voters to state a positive preference, conventional methods conflate genuine intent with social desirability bias. The result isn't signal — it's systematically biased estimates.
Reweighting algorithms cannot manufacture clarity from fundamentally ambiguous behavioral data. In a fragmented political landscape, asking "Who will you vote for?" is the wrong question.
The Pivot
The Negative Partisanship Approach
Decades of political science research reveal a counterintuitive reality: voter behavior is increasingly driven not by enthusiasm for a candidate, but by hostility toward the alternative.
VoxSim applies MaxDiff (best-worst) scaling to LLM synthetic agents, asking a more stable, honest question:
"Who would you never vote for?"
By mapping the exclusion boundaries of the electorate, the ambiguous "undecided" segment collapses. What remains is a sharply defined electoral universe for each candidate, ready for strategic targeting.
Policy Wargaming over Polling
Three Operational Contexts
Second-Round
Feasibility Mapping
Map the true size of a candidate's accessible electoral universe before persuasion resources are deployed. Know who is mathematically in play.
Crisis
Stress-Testing
Simulate how a judicial decision, economic shock, or scandal will impact exclusion rates across 12 distinct clusters. Test message variants before public deployment.
Cross-Ecosystem
Resonance Analysis
Identify messages that successfully persuade a target demographic while avoiding unintended identity-protective backlashes in adjacent clusters.
Validated in Volatility
The World's Most Analytically Demanding Electorate
Developed and stress-tested in Turkey's 12-cluster, 312-cell electorate across 26 NUTS-2 regions. High polarization, rapid agenda shifts, fragmented opposition, dual-round presidential dynamics. If the simulation holds here, it scales anywhere.
We tell you where to look,
not what you will find.
VoxSim does not predict static vote shares. It generates actionable hypotheses. Our outputs are directional fault-line maps that narrow your hypothesis space, making subsequent fieldwork hyper-efficient.
We treat transparency as a feature, not a legal requirement. Known systematic deviations — including LLM WEIRD-bias affecting conservative segment representation — are explicitly documented, quantitatively offset, and available upon request.
Technical Whitepaper
Negative Partisanship Analysis via LLM Synthetic Agents
Full methodology, calibration framework, known bias documentation, and use-case boundaries. Suitable for technical due diligence.
Designed for Decision-Makers
Structured Engagement Tiers
-
WS
Workshop
For independent strategists and boutique consultancies.
-
FD
Field
For campaign teams executing rapid-response operations.
-
HQ
Headquarters
For national parties and institutional clients requiring multi-module access.
Request a Strategy Session
We'll walk through your specific political context, identify the most relevant clusters, and scope a simulation that answers the question your team is actually asking.
- Full technical documentation on request
- Calibration methodology & bias audit reports
- EU AI Act Annex III compliance documentation