llm-judge

To evaluate if your system meets requirements, you provide the requirement list and relevant design or code materials to generate a structured JudgeReport. This report scores requirements fit, identifies missing items, and highlights implementation risks based strictly on your evidence.

Evidence-based scoring in system evaluation assesses compliance by matching provided artifacts directly against your stated requirements. It produces a structured JSON report marking items as satisfied, partially satisfied, or missing, ensuring the verdict relies on materials rather than redesigns.

Yes, you can check multi-agent communication logic and interface consistency. The evaluation verifies logical consistency, component responsibility coverage, and communication correctness across mixed design-and-code systems to identify inconsistencies against your requirements.

To assess code quality and security risks, include your code artifacts alongside the requirements list. The evaluation applies security and code-quality checks guided by known risk categories and confidence thresholds, reporting implementation risks in the final JudgeReport.

No, the requirements analysis does not propose redesigns or regenerate artifacts. It provides an independent evaluation that explicitly marks satisfied, partially satisfied, or missing items, focusing strictly on compliance assessment rather than solution generation.

To generate a structured JudgeReport, you need to provide a list of system requirements and the relevant conceptual design or code artifacts. The evaluation uses these materials to determine overall compliance and output the JSON report with requirements fit scoring.