---
license: cc-by-4.0
language:
- en
tags:
- school
- psychology
---

# MTSS School-Based Behavioral Health QA Dataset

This dataset contains a curated collection of real-world questions and corresponding expert responses related to Multi-Tiered Systems of Support (MTSS) and school-based behavioral health integration. Questions vary in style—ranging from open-ended reflections to specific, context-based prompts—and are often rooted in realistic school scenarios, with some referencing external case files.

## Dataset Summary

The dataset is designed to evaluate large language models (LLMs) on nuanced, domain-specific question-answering tasks. It captures authentic school-based inquiries around integrating behavioral health practices into academic systems.

Each entry includes:
- A **question** posed by a school professional (e.g., administrator, counselor, MTSS specialist)
- A **gold standard answer**, written by subject matter experts
- The **role** of the participant
- A reference to **case files** (when applicable)

This dataset enables fine-grained assessment of LLM performance on educational, behavioral health, and systems-level implementation topics.

---

## Dataset Structure

| Column        | Description                                                                 |
|---------------|-----------------------------------------------------------------------------|
| `questions`   | Real-world queries about MTSS and school-based behavioral health practices  |
| `gold_answers`| Human-validated answers providing expert guidance or reflection             |
| `roles`       | Role of the person involved (e.g., `School_Admin`, `Counselor`, etc.)       |
| `case_files`  | Linked case information or placeholder (`none` if not applicable)           |

---

## Example Entry

```json
{
  "questions": "We are just starting with MTSS. What are the core features of the framework? Where should we begin?",
  "gold_answers": "Core features: Effective teams, using data... (full structured response)",
  "roles": "School_Admin",
  "case_files": "none"
}