synthetic-workflow-polish-eval

To evaluate synthetic workflow polish for demo readiness, grade step outputs against a standardized 7-dimension weighted rubric covering narrative coherence, patch quality, smoke render success, and visual fitness. This generates a structured verdict YAML with calibration metadata.

Rubric scoring for visual judging evaluates rendered workflow screenshots against standardized criteria like visual hierarchy and brand fit. It uses a visual judge to capture out-of-chain visual fitness, ensuring polished workflows meet strict stakeholder-ready demo standards.

Enforce hard blockers for broken smoke renders and mismatched FLW names by automatically failing evaluations when critical issues are detected. This prevents low-quality demos from progressing through the synthetic workflow polish stage.

Yes, you can generate a structured YAML verdict report for patch conformance and domain language fit. This report includes dimension scores, hard deduct triggers, and calibration metadata to support continuous rubric improvement.

Automated polish evaluation relies on capturing rendered screenshots for visual judging and enforcing hard blockers for mismatched FLW names or blocked visual judge verdicts. It cannot assess subjective design aesthetics beyond the defined 7-dimension weighted rubric.

Yes, synthetic workflow polish evaluation operates without external dependencies. It internally evaluates text-based patch conformance and uses a visual judge to score out-of-chain visual fitness from captured screenshots to produce a standardized verdict.