Dynamic Hazard Rate BART

A configurable, offline desktop instrument for the Balloon Analogue Risk Task.

Dynamic Hazard Rate BART lets a research lab administer, score, and export Balloon Analogue Risk Task (BART) sessions without writing code and without a network connection. A researcher designs a study in a point-and-click Study Setup screen — choosing from a curated library of eleven hazard families with a live expected-value preview — and runs participants through a conventional BART flow. Sessions are scored locally by the bundled Python engine and written to per-session files plus a shared Master CSV ready for SPSS or R.

The flagship paradigm is the dynamic hazard rate: at each pump \(k\) the balloon bursts with probability \(k / N\) (a sequence of independent trials with a linearly increasing hazard). This moves the expected-value (EV) optimal stopping point from the unattainable \(N/2\) of the classic task down to roughly \(\sqrt{N}\) — a target participants can actually reach — so the task distinguishes cautious, calibrated, and reckless play rather than collapsing into a one-dimensional proxy for raw risk exposure. The classic uniform (Lejuez) model and nine other hazard families are available as configuration choices for baselines and replications.

The project comprises:

• the standalone desktop instrument (Tauri shell + React/Vite UI + a loopback-only Python scoring sidecar; see the researcher quickstart),

• a reusable Python scoring engine (scoring/) that derives 40+ psychometric metrics from raw event telemetry, and

• Monte Carlo verification tooling that confirms every configuration’s numerically computed EV optima (hazard-family reference).

It was developed as the measurement instrument for a research program at the Middle East Technical University (METU).

Where to start

Running a study? Follow the researcher quickstart, then see Data Outputs for what lands in your output directory. Using the engine as a library? Read Task Design for the model, then Quick Start to score your first session. Looking for a specific number? Jump to the Metrics Reference.

The default study

Out of the box the instrument loads the validated three-color linear-hazard design — 30 balloons (10 per color) in shuffled order, deliberately neutral colors (no red/green danger framing), $0.25 per banked pump. Every element below is configurable in Study Setup.

Color	Max pumps (\(N\))	Risk tier	EV-optimal stop (\(s^*\))	Peak EV	\(P(\text{survive } s^*)\)
Purple	128	Low	11	6.46	0.59
Teal	32	Medium	5	3.04	0.61
Orange	8	High	2	1.31	0.66

Contents

For researchers

• Researcher quickstart
  • 1. Download
  • 2. Install
  • 3. Create a study
  • 4. Run participants
  • 5. Collect data
  • 6. Customize
  • 7. Troubleshooting
• SmartScreen & antivirus
  • What is SmartScreen?
  • Why the warning appears here
  • How to bypass it (researchers)
  • IT administrator notes (managed lab machines)
  • Antivirus false positives
  • Future: code signing
• Kiosk mode & OS lockdown
  • The in-app lock (built in)
  • A true OS kiosk on Windows (Assigned Access)
  • macOS
• Hazard-Family Reference
  • Simulation-verified optima
  • The families
  • Choosing a family
• Data Outputs Reference
  • The output directory
  • Study-level provenance files
  • Session files
  • Test Run (practice) sessions
  • The Master CSV
  • The Trials CSV

The engine as a library

• Installation
  • Requirements
  • From source
  • Verifying the install
  • The game client
  • Building the documentation locally
• Quick Start
  • 1. Build an event log
  • 2. Score it
  • 3. Validate before scoring (optional)
  • 4. Reading the result

The task & engine

• Task Design
  • The problem with the standard BART
  • The dynamic-hazard model
  • Why the optimum is \(\sqrt{N}\)
  • Discrete optima
  • Monte Carlo verification
  • The three-color profiles
  • Reward structure
• The Scoring Engine
  • Pipeline overview
  • The censoring correction (collected-only metrics)
  • Calibration metrics (the sequential EV core)
  • Learning and adaptation
  • Consistency and timing
  • The composite score
  • Flat-strategy detection
  • Behavioral profiles
• Session Validation
  • What validate_bart_session returns
  • Checks
  • Auto-repeat detection
  • The RNG fallback
  • Interpreting validity
• Metrics Reference
  • Volume & outcome
  • Calibration (sequential EV)
  • Learning & adaptation
  • Behavioral indices
  • Composite, profile & validity
  • Per-color metrics

Components & tooling

• Game Client
  • Component contract
  • Session generation
  • The explosion model
  • Event logging
  • Submission
• Data Schemas
  • Events: the atomic unit
  • Sessions
  • Event validation
  • Output models
• Scripts & Verification Tooling
  • Optima verification (all hazard families)
  • Hazard-family EV figure
  • Default-study EV figures
  • Synthetic data generation

Reference

• API Reference
  • Scoring engine
  • Data schemas
  • Event validators
• References
  • The BART and sequential-risk modeling
  • The attitude–behavior gap and risk preference
  • Software
  • Companion research report

Citing this software

If you use this instrument in your research, please cite the archived release (see CITATION.cff):

Yılmaz, A. S. (2026). Dynamic Hazard Rate BART: A Configurable Offline Desktop Instrument for the Balloon Analogue Risk Task (Version 1.1.0) [Computer software]. https://doi.org/10.5281/zenodo.20592164

License

Released under the MIT License.