Methodology
How it
works
You are not looking at survey data. You are interacting with a population — structured, calibrated, and capable of revealing how reactions form before they become visible.
Architecture
Four layers.
One living model of a population.
Each layer builds on the previous. The result is not a segmentation framework or a survey tool. It is a model of how real people receive, process, and propagate information — built from the ground up on empirical data.
Foundation
01
Synthetic Population
Real respondents, unified across 15 national datasets spanning 2020–2026. Structurally aligned with the 2021 Census through iterative proportional fitting. Not simulated personas — not algorithmic agents — actual people with actual attitudinal profiles, calibrated to represent a national population of 14,536.
What this means in practice: Every individual in the model is a real survey respondent. Demographic imbalances across datasets are corrected through IPF calibration. The result is a population that is both empirically grounded and structurally representative.
Why it matters: Simulated or synthetic personas introduce researcher assumptions. Our population inherits real complexity — including the contradictions, weak correlations, and structural independence between attitudinal dimensions that only real data reveals.
14,536 real respondents · Census-calibrated
Behavioral model
02
Agentic Sandbox
Each individual becomes an active agent with a full behavioral profile. The sandbox simulates a social network of 91,978 edges. Reactions are inferred from structural dispositions and social position — not generated by a language model, not extrapolated from averages.
How it works: Messages, events, and scenarios are introduced into the network. Each agent responds according to its latitude of acceptance — the range within which a message is absorbed, processed, or rejected. Propagation follows the actual topology of social connections.
Key insight from simulation: Direct counter-messaging produces a boomerang effect in all 12 typologies. The model identifies not just who agrees — but who resists, who amplifies, and who reacts in the opposite direction.
Monte Carlo simulation · 91,978 network edges
Attitudinal structure
03
Pre-cognitive and Cognitive Layers
Behavior emerges from the interaction of two distinct layers. The pre-cognitive layer captures stable underlying dispositions — structural scepticism, institutional legitimacy, pragmatic fatalism — that operate before conscious evaluation. The cognitive layer captures explicit beliefs and active reasoning.
Pre-cognitive layer: Five independent dimensions (D1–D5), each decomposed into 2–4 sub-dimensions. These are not opinions — they are dispositions that determine how any message is received before the person is aware of receiving it.
The key finding: The five dimensions are almost entirely uncorrelated. There is no coherent ideological profile that predicts behaviour across dimensions. The same person can score high on conspiracism and high on institutional trust — these are separate activation systems.
5 dimensions · 12 sub-dimensions · calibrated per segment
Where reactions form
04
Expectation Layer
Expectations about events, others, and perceived norms. This is where reactions are actually formed — not in response to what is, but in response to what people believe others will think. The third-person effect is not a bias to correct. It is the mechanism to measure.
Third-person effect: Across all 12 typologies, people consistently perceive others as more susceptible to messages than themselves. The gap is measurable, stable, and larger for disinformation than for health information. It is universal — and activatable.
What this enables: Messages calibrated to the perception gap — not to stated opinions — move through the network differently. This is the layer that makes anticipation possible: not forecasting what people believe, but forecasting what they will believe others believe.
Perception gaps · third-person effect · expectation modelling
Capabilities
What you can do
Test a scenario — How would people react if X happens? Run events, messages, or policy decisions through a living population model before they occur.
Explore segments — Who reacts, who amplifies, who resists? Understand not just the average response but the structural distribution of reaction types.
Track perception gaps — What do people believe versus what they think others believe? Map the distance between personal position and perceived social norm.
Identify the nudge — Which message enters the latitude of acceptance for each typology? Which triggers a boomerang effect? Test before deploying.
Map mobilisation potential — Who is passively aligned versus actively mobilisable? The gap between acceptance and action is measurable in the data.
What you see
Perception gaps (Qa–Qb) — the distance between what people believe and what they think others believe, per segment.
Reaction distributions — not average response, but the full distribution across typologies, weighted by population share.
Segment-level dynamics — which typologies assimilate, which elaborate, which produce backfire. At what intensity, through which channels.
Propagation paths — how a signal travels through the social network, which clusters amplify it, where it saturates, and where it dies.
Emerging patterns — structural shifts in attitudinal anchors before they become visible in conventional polling or media monitoring.
Output types
Reaction distributions
Full distribution of responses across all 12 typologies, weighted by population share. Not an average — a structural map.
Perception gaps
Quantified distance between personal belief and perceived social norm. The engine of social change, made visible.
Segment dynamics
Assimilation, elaboration, and boomerang effects per cluster. Who amplifies, who resists, at what threshold.
Emerging patterns
Structural shifts in attitudinal anchors — identified before they surface in polling or media monitoring.
You're not observing opinions.
You're observing how
reactions form.
A predictive sandbox built on real population data. Not a survey. Not a panel. A living model of how a society receives, processes, and propagates information.