---
language:
- en
- zh
license: apache-2.0
task_categories:
- text-classification
- feature-extraction
- zero-shot-classification
tags:
- agent
- decision-making
- cross-encoder
- reranker
- modernbert
- tool-use
size_categories:
- 10K<n<100K
---

# Agent-Decision-30k: Multi-Task Operational Decisions Dataset

**Agent-Decision-30k** is a curated dataset of 30,456 high-fidelity operational decisions for training and benchmarking System-1 (fast-thinking) decision engines and cross-encoders.

## Dataset Structure

Each sample is formatted as a JSON object:
```json
{
  "context": "Detailed observation, system logs, environment state, or DOM context",
  "instruction": "Target goal, user intent, or operational constraint",
  "candidates": {
    "action_a": "First candidate option",
    "action_b": "Second candidate option",
    "action_c": "Third candidate option"
  },
  "gold": "action_a",
  "domain": "browser_dom | api_dispatch | ecom_brand | ecom_specs | ecom_category | legal_clause"
}
```

## Domains Covered

1. **Browser DOM Automation**: XPath selection, button clicking, form field focus.
2. **API Tool Routing**: Function calling and handler dispatch under strict argument constraints.
3. **E-Commerce Brand Extraction**: Canonical brand attribution across messy product titles.
4. **E-Commerce Specs & Attributes**: Multi-attribute constraint satisfaction.
5. **E-Commerce Category Prediction**: Hierarchical taxonomic placement.
6. **Legal & Contract Alignment**: Clause matching and regulatory compliance verification.

## Intended Use
Designed to train sub-15ms cross-encoders (such as [FastDecider-149M](https://huggingface.co/mkzero/FastDecider-149M)) for reliable agent dispatch.