Prompt Benchmark

# Prompt Scoring Benchmark

These examples calibrate production-readiness scoring. Ranges allow for policy
selection and reviewer judgment; a higher tier should consistently outscore the
tier below it.

| Tier | Example prompt | Policy Engine | AI Reviewer | Hybrid |
|---|---|---:|---:|---:|
| Very Poor | `Tell me about AI.` | 0–20 | 10–25 | 4–22 |
| Poor | `Summarize this.` | 15–30 | 25–45 | 19–36 |
| Average | `Summarize the supplied article for a general audience. Return one short paragraph covering the main argument and key evidence.` | 50–70 | 55–70 | 52–70 |
| Good | `Summarize the article between <article> tags for an executive audience. Return five bullets: thesis, three supporting facts, and conclusion. Use only the article and say when evidence is missing.` | 75–90 | 70–85 | 73–88 |
| Excellent | `You are an expert executive communications editor. Summarize the supplied article between <article> tags for an executive audience. Return exactly five bullet points only, each under 20 words. Use only the article; do not invent facts. If a fact is missing, write "Not stated." Return only the bullets with no extra commentary.` | 90–95 | 90–95 | 90–95 |

Hybrid uses the default 60% Policy Engine and 40% AI Reviewer weighting. If the
Policy Engine score is below 20, the Hybrid score cannot exceed 40. Blocking
policy failures prevent a Ready decision regardless of the AI score.