coreason-assay 0.4.0

Quality Control (QC) laboratory of the CoReason platform

coreason-assay

The Scientific Testing Engine for AI Agents.

coreason-assay is the Quality Control (QC) laboratory of the CoReason platform. It provides a rigorous framework for evaluating the performance, safety, and alignment of AI agents before they are deployed to production.

Features

• Benchmark Evaluation Corpus (BEC) Management: Easily ingest test cases from CSV, JSONL, or ZIP archives.
• Simulation: Run agents in a controlled sandbox with mocked tools and injected context.
• Glass Box Grading: Evaluate not just the answer, but the reasoning process (Faithfulness, Alignment, Tone).
• Report Cards: Generate detailed reports with drift detection and pass/fail metrics.

Quick Start

Installation

poetry install

Usage

Run the CLI to upload a test corpus:

poetry run coreason-assay upload path/to/bec_archive.zip

Server Mode (Service C)

coreason-assay can also be run as a containerized microservice exposing a REST API.

# Start the server (requires Docker)
docker build -t coreason-assay:0.4.0 .
docker run -p 8000:8000 coreason-assay:0.4.0

The service exposes endpoints for uploading corpora (POST /upload) and running assays (POST /run). See docs/usage.md for details.

Documentation

For full documentation, including architecture details, usage guides, and examples, please visit the docs folder.

• Architecture
• Usage Guide
• Vignette
• Product Requirements

License

This software is proprietary and dual-licensed. See LICENSE for details.