Dataset Card for Crisis-QA

Dataset Details

Crisis-QA is a large-scale synthetic benchmark designed to evaluate the Causal Reasoning and Physical Validity of Artificial Intelligence models in the domain of National Critical Infrastructure. Unlike general-purpose QA datasets, Crisis-QA focuses on "high-stakes" scenarios where a hallucination (factually incorrect generation) could lead to loss of life or economic damage.

The dataset contains 1,926 unique scenarios across four critical domains:

1. Wildfire Physics: Testing understanding of rate-of-spread given temperature, wind, and humidity.
2. Hydrology (Floods): Testing understanding of water velocity, distance, and travel time.
3. Structural Integrity: Testing safety assessments of bridges and infrastructure under load.
4. Supply Chain Logistics: Testing routing decisions under adverse weather conditions.

• Curated by: Annesha Chowdhury
• Language: English
• License: MIT
• Task Categories: Question Answering, Fact Verification, Causal Inference

Dataset Description

The integration of Large Language Models (LLMs) into emergency management systems presents a "Trust Gap." While models can generate plausible-sounding reports, they often fail to adhere to physical laws (e.g., predicting a flood arrives instantly over 100 miles). Crisis-QA addresses this by providing a rigorous test set of physics-informed questions.

Each entry in the dataset consists of:

• Scenario Context: A set of environmental variables (e.g., "Wind Speed: 60mph", "River Depth: 15ft").
• The Claim/Prediction: An assertion made by an AI or social media user (e.g., "The fire is contained").
• The Evaluation: A ground-truth verification based on physics (e.g., "False. 60mph winds make containment impossible").
• Distractors: Plausible but incorrect reasoning paths to test the model's logical robustness.

This dataset serves as a benchmark for the Autonomous Verification Architecture (AVA), a framework for building "Digital Immune Systems" for public safety.

Dataset Sources

• Methodology: The scenarios were generated using a combinatorial physics engine that permutes environmental variables (Temperature, Wind, Load, Material) against known physical constants (e.g., Manning's Equation for water flow, Rothermel's Model for fire spread).

Dataset Card Contact

Maintainer: Annesha Chowdhury
Maintainer Email: anneshachowdhury71@gmail.com Affiliation: Crisisense / Independent Researcher
Profile: https://huggingface.co/AnneshaChowdhury

---

Citation

If you use this dataset in your research, please cite:

@dataset{chowdhury2026crisisqa, author = {Chowdhury, Annesha}, title = {Crisis-QA: A Benchmark for Causal Verification in Critical Infrastructure}, year = {2026}, publisher = {Hugging Face}, url = {https://huggingface.co/datasets/AnneshaChowdhury/Crisis-QA} }