sadd:judge

To evaluate work artifacts with tailored rubrics, you initiate a two-phase evaluation pipeline that generates custom scoring criteria and applies them with isolated context to produce a structured, evidence-backed report without modifying the original work.

A two-phase meta-judge evaluation pipeline is a process where a meta-judge first extracts context to produce a YAML evaluation specification, followed by a judge applying those tailored criteria with fresh context to generate an evidence-backed report.

You generate custom evaluation criteria by providing work context to the system, which dispatches a meta-judge to produce a tailored YAML evaluation specification that defines the rubrics and scoring rules for the artifact.

The evaluation process does not modify the original work, because the system enforces strict context isolation during the evaluation phases and validates the results before presenting a structured report.

Yes, you can use isolated context to prevent bias, because the system applies freshly generated rubrics to the work artifact with strict context isolation and enforces evidence requirements for every score to mitigate bias.

A limitation of automated rubric-based artifact evaluation is that it requires sufficient work context for the meta-judge to generate meaningful criteria, and the judge must enforce strict evidence citations for every score to ensure validity.