goalchecklisted

# Goalcheck: CTO-Conversion-Driven Iteration Loop **Target file:** `$ARGUMENTS[0]` **Theme:** `$ARGUMENTS[1]` (a=CTO, b=Editorial, c=Epistemic, d=Curriculum, e=Retrieval) **Threshold:** 3.5 average across all evaluators (judges score 1-5) **Max iterations:** 4 ## Strategic Frame The goal is NOT "pass the eval." The goal is: **minimum content that makes a CTO contact Antti and share the site internally.** The eval is a quality gate — it catches epistemic problems that would make a CTO dismiss the content. But the DIRECTION of improvements is conversion-driven: 1. **30 seconds:** Does the CTO understand what this is and who it's for? 2. **2 minutes:** Do they believe this person is credible (practitioner, not consultant)? 3. **Share trigger:** Is there something worth sending to their team? 4. **Contact trigger:** Do they see thinking they want access to? 5. **Next step:** Is the CTA obvious? Every improvement must serve one of these five questions. If it only serves the eval score, it's wrong. ## Protocol You are the orchestrator. **Preserve your context window.** Delegate file reading, editing, and eval running to subagents. You read only the iteration log and short summaries. ### Judge files | Theme | Judge file | Evaluators | Score | |-------|-----------|------------|-------| | A: CTO Prompt | `evals/judges/cto-prompt-judges.md` | Relevance, Specificity, Actionability, Evidence Grounding | 1-5 each | | B: Editorial | `evals/judges/editorial-judges.md` | Source Ver