discover-confidence

To score a blueprint for quality assurance, you need a Python runtime, a rubric loaded from project templates, and the blueprint markdown file on disk. The process evaluates research coverage, reference citations, completeness, and structural smells to produce a reproducible score.

Yes, you can evaluate blueprint completeness without model calls. This deterministic scoring mechanism operates independently of model inference, using a fixed rubric loaded from templates to assess structural smells and coverage, ensuring a reproducible score report every time.

Deterministic blueprint evaluation outputs a JSON score report containing per-dimension scores, a final score, a verdict, and hard-cap details. It assesses research coverage, reference citations, blueprint completeness, and structural smells to provide transparent quality judgments.

Blueprint scoring requires a Python runtime and the pyyaml dependency to load the rubric configuration. You also need the blueprint markdown file saved on disk, allowing the deterministic evaluation scripts to parse and score the document locally.

A deterministic rubric ensures consistent, auditable quality judgments for blueprint reviews. Unlike model-based evaluation, it provides a reproducible score by applying fixed criteria to research coverage and structural smells, eliminating variance in quality assurance workflows.

You integrate deterministic blueprint scoring by running it within the discovery workflow after generating a blueprint. It assesses the blueprint against a rubric to produce a score report, which guides merges and improvements during the discovery process.