Quality audit: onboarding-analysis skill

Run the onboarding-analysis skill end-to-end (Step 0 through Step 4) and evaluate output quality. Check: 1. Step 0 (Redshift fetch) — do all 5 queries return reasonable data? Row counts consistent with expectations? 2. Step 1 (data scripts) — do all 5 scripts run without error? Output numbers internally consistent? 3. Step 1.25 (top-of-funnel) — coverage, plan-select conversion, program type splits look reasonable? 4. Step 1.5 (archetypes) — does inline Python run without index alignment errors? Are archetypes meaningful? 5. Steps 2–2.7 (campaign analysis, verification, investigations) — do findings hold up on closer inspection? 6. Step 3 (report.md) — follows template structure? Writing rules met? No raw field names? Denominators present? Caveats included? 7. Step 4 (HTML outputs) — explorer.html and story.html render correctly? Charts populate? Data matches report? Grade each step: Pass / Minor issues / Major issues. File the report in the output directory.