End-to-end test: affiliate-onboarding-pulse — run full skill, audit output quality

Run affiliate-onboarding-pulse end-to-end (Step 0 through Step 6) and critically evaluate the OUTPUT. ## What to test 1. Run the full skill from Step 0 (Redshift fetch) through Step 6 (today.md) 2. Read pulse.md as if you're the Affiliates PM checking health 3. Read today.md as if you're a CSE looking for affiliate merchants to help ## What to audit (output quality) ### Usefulness - Does the two-tier stall detection (pre-entry vs post-entry) make sense to a reader who doesn't know the affiliate funnel? - Are the "Reading the table" guides actually helpful for CSE outreach? - Does the pulse distinguish meaningfully from the referral pulse, or could they be merged? - Is the AAM rate context useful as reference-only, or does it need more framing? ### Clarity & Readability - Is the pulse concise enough for a quick scan? - Do the pre-entry and post-entry tables have too many columns? - Does the Narrative feel natural or templated? - Are the stall point descriptions in plain enough language? ### Template prescriptiveness check - Same lens as the other two: identify forced structure that hurts readability - The affiliate funnel is more complex than referral — does the template handle that complexity gracefully or does it create information overload? - Does the "came back after Nd" / "ran wizard Nx" notation work in practice? ## Output Write findings alongside the generated pulse output. Flag specific passages, suggest rewrites. Recommend SKILL.md template changes if issues are systemic.