discover-plan-confidence

You can score a discovery plan by running it through deterministic evaluators for research coverage, reference citations, plan completeness, and spec smells to generate a weighted final verdict.

Deterministic plan gating is the process of evaluating a plan's structural quality using zero-LLM checks to provide fast feedback and clear hard caps before allowing the plan to proceed to execution.

The M2 rubric evaluates research coverage and reference citations by running independent checks that assess structural completeness, returning a weighted score and detailed per-dimension JSON report.

Yes, you can obtain a structured JSON score report containing the final_score, final_verdict, and per-dimension details by applying the scoring checks to a valid discovery plan slug.

You need the pyyaml dependency installed in your environment to run the Python scripts that execute the deterministic scoring checks and generate the final verdict.

The zero-LLM scoring approach relies entirely on deterministic structural checks, meaning it cannot assess semantic nuances or logical validity outside the M2 rubric's defined hard caps and spec smell rules.