Documentation

A reference for how Behavioural Engines works: its models, workflow, features, and limits.

Overview

Behavioural Engines runs the standard models of behavioural economics, including Prospect Theory, Hyperbolic Discounting, and Rational Inattention, in the browser. You set the parameters, run a deterministic simulation, compare results, and export the output. You can also collect real responses through shareable surveys and fit model parameters to them.

Every model is implemented from a peer-reviewed paper, with the full citation, the equations, and an explanation of each parameter.

Purpose

The standard models of behavioural science live in papers and textbooks. To run one you normally have to write code yourself or work within whatever a statistics package supports. Behavioural Engines puts working implementations of these models in one place in the browser.

Each model is implemented from its source paper with the original mathematics, given parameter controls with sensible ranges, and connected to charts so you can see what changes when you change an input.

Workflow

Select a model

Choose a model in the workspace. You can switch between models with the top-bar tabs or the ⌘K command palette.

Configure parameters

Adjust the core and advanced coefficients, load a saved preset, or start from one of the pre-built scenarios.

Run the simulation

Press Run Simulation (or ⌘↵). The backend computes the result deterministically and returns the metrics and chart data.

Compare & explore

Pin a result as a baseline, change parameters, and run again. Both results appear side by side in the analysis panel and as overlaid lines on every chart.

Review & export

Every run is saved to your history with a searchable name. Pro users can re-run past simulations, copy shareable links, and export results as PDF, CSV, or JSON.

Run an experiment

Click Run Experiment on a supported model and share the generated link. Fitted parameters appear on the Experiments page as responses arrive.

Models

Structural Adoption

Applied Microeconomics

Logistic structural model estimating bounded adoption probability from behavioural levers with diminishing incentive returns.

Prospect Theory

Behavioural Economics

Non-linear value model with probability weighting and loss aversion.

Hyperbolic Discounting

Behavioural Economics

Comparative intertemporal choice model contrasting hyperbolic and exponential time discounting.

Expected Utility Theory

Classical Economics

Normative decision model computing expected utility under constant relative risk aversion (CRRA).

Inequity Aversion (Fehr–Schmidt)

Behavioural Economics

Structural social-preference model where utility depends on one's own payoff and on payoff differences relative to others.

Rational Inattention

Behavioural Economics

Decision model where agents optimally limit attention due to costly information processing.

Regret Theory

Behavioural Economics

Counterfactual utility model where outcomes are evaluated against what the foregone alternative would have given in the same state.

Platform Features

Workspace

Parameter control

Every coefficient in every model is adjustable. Core inputs are always visible, and advanced options sit behind a collapsible panel.

Parameter presets

Save named configurations per model. Presets sync across devices for signed-in users via server-side storage.

Scenario library

Each model comes with pre-built scenarios based on real-world situations, which give you a working configuration to start from.

Command palette

Press ⌘K (or Ctrl+K) to open a searchable command palette. Navigate to any model or page without touching the mouse.

Simulation & Analysis

Comparison mode

Pin any result as a baseline, then adjust parameters and re-run. Metric panels show both side-by-side; charts overlay dashed baseline lines against solid current lines.

Visual charts

Every model renders its curves as interactive charts: value functions, discount curves, probability weighting, and the rest.

Live analysis

Results appear in a structured analysis panel with clearly labelled metrics, colour-coded by sign, immediately after each run.

History

Simulation logs

Every run is stored with its full parameters, results, model version, and timestamp. Paginated and searchable.

Named runs

Give any simulation run a custom name directly from the History page or the detail view. Names sync across devices.

Re-run

Load any past simulation's parameters back into the workspace with one click and re-run with modifications.

Experiments

Shareable surveys

From any supported model, click Run Experiment to generate a shareable link. Respondents answer a short series of either/or choice questions without creating an account, and each receives a personalised reading of their results.

Fitted parameters

Each response is converted into fitted model parameters by revealed-preference estimation. The experiment page shows per-parameter distributions with mean, standard deviation, median, and range.

Pause & resume collection

Pause an experiment to stop accepting responses, and resume it at any time. While paused, the survey link shows a closed notice. Plus accounts can hold 3 experiments; Pro and Lifetime have no limit.

CSV export

Download every respondent's fitted parameters, submission date, and duplicate-network flag as CSV for analysis in Excel, Python, R, or SPSS.

Exports & Sharing

Export as PDF

Generate a formatted PDF of a simulation, including its parameters, results, and metadata.

Export as CSV

Download structured tabular output for further analysis in Excel, Python, R, or any data tool.

Export as JSON

Download the full simulation object as JSON for use in your own pipelines.

Shareable links

Copy a direct link to any saved simulation run and share it with colleagues. The link preserves parameters and results.

Experiment Methodology

Experiments measure behavioural parameters from real choices. Respondents answer a series of binary questions, and the platform estimates the parameter values that best explain their answers by grid search.

Task battery

Each experiment presents a fixed battery of calibrated either/or choice tasks. For Prospect Theory this is 13 tasks: five mixed gambles against a certain £0 (identifying loss aversion λ), four gain gambles at small and large probabilities (identifying probability weighting γ), and four 50/50 gain gambles against certain amounts (identifying value curvature α). Each task is constructed so its indifference point sits at a known parameter crossover, meaning every answer rules parameter regions in or out.

Estimation

Fitted parameters are found by exhaustive grid search over 512 candidate (α, λ, γ) triples, selecting the combination that correctly predicts the largest number of the respondent's observed choices under the model's value and probability-weighting functions. Ties resolve towards the Kahneman–Tversky empirical medians (α = 0.88, λ = 2.25, γ = 0.61), so an uninformative response pattern returns population-typical values.

Data quality

One response is accepted per device via an idempotent respondent token, and responses sharing a network are flagged (not excluded) in the researcher's results, since shared networks are common in classroom settings. Raw choice data and fitted parameters are exportable as CSV for external analysis.

Limitations

Grid-search estimation returns the best point on a discrete grid. It does not produce continuous maximum-likelihood estimates with standard errors. With a short series of binary choices, the data can be consistent with several parameter combinations, so results should be read as behavioural profiles and not as precise psychometric measurements. Experiments are currently available for Prospect Theory and Hyperbolic Discounting, with further models planned.

Boundaries & Assumptions

Understanding what this platform is not:

Not a forecasting engine for real populations or empirical datasets.
Not a substitute for field trials, experiments, or empirical calibration.
Not a claim that human behaviour is fully captured by any single model.
Outputs are structural and deterministic. They reflect the model, not observed data.
Parameters are illustrative unless calibrated against real measurements.

Real behaviour is probabilistic even though these models are deterministic. Incentives act through perception as well as objective value. Friction reduces action even when incentives are positive. Social adoption can amplify small initial differences. Every output reflects the model structure and is not an empirical claim about any population.