Core concepts

Datasets & prompts

See which data each evaluation module uses and the current requirements for uploading your own test data.

Each evaluation module uses data suited to its risk question. This page gives a high-level map and documents the user-upload format currently supported by the platform.

Platform datasets

The dataset depends on the evaluation module. Open the module guide for its detailed coverage, generation method, and interpretation.

Benchmark testing

BenchDX datasets

Uses platform benchmark cases organized by selected safety categories or standards.

Robustness testing

RobustDX data

Uses uploaded seed prompts and selected attack methods to generate adversarial test prompts.

Coming soon

HalluDX data

Planned data covers selected domains, topics, questions, and repeated responses.

Coming soon

AlignDX data

Planned data begins with policy or regulatory sources and derived test scenarios.

Upload your own data

User-uploaded data is currently supported when creating a RobustDX evaluation.

Requirement	Current rule
Evaluation module	RobustDX
Accepted format	One .csv or .txt file
Data layout	Place each seed prompt in a separate row or cell
Before continuing	Check the detected prompt count and preview