# STAR Stories — Master List

Short, plain-language interview stories. Each one maps to a common behavioral question. Claude Code reads this file alongside the JD and the super resume when building interview-prep material.

## 1. The vanity metric that almost shipped (failure, learning)

- **Situation:** I was leading an AI-search optimization effort for a product docs site. We wanted the docs to work well for AI agents, not just human readers.
- **Task:** Show measurable progress on "agent friendliness" so the team could prioritize doc changes.
- **Action:** I started with an agent-friendliness score and chased it for two sprints before realizing it measured writing style, not outcomes. I stopped, owned the miss, and built an eval harness that measured whether agents could actually complete real tasks using the docs.
- **Result:** The eval caught regressions the score had been hiding, and it became the team's gate for doc changes. The failure taught me to ask "what decision does this metric drive" before adopting one.

## 2. Docs serve two goals (receiving tough feedback)

- **Situation:** I reorganized an AI-search doc set by feature complexity, simple concepts first, advanced ones later.
- **Task:** Get the information architecture approved by product.
- **Action:** My PM pushed back hard: new customers learning new features and existing customers maintaining old ones are equally important audiences, and my structure served only the first. Instead of defending the draft, I asked her to walk me through the second audience's top tasks, then reframed the architecture around both entry points.
- **Result:** The revised structure shipped, and "who is this page's second reader" became a question I now ask on every doc plan.

## 3. Data Management information architecture at Splunk (scale, leadership)

- **Situation:** Splunk's Data Management docs had grown product-by-product, so users had to already know the product name to find anything.
- **Task:** As lead writer, redesign the IA so users could navigate by what they were trying to do.
- **Action:** I ran a content audit across the product family, mapped docs to user jobs instead of product SKUs, and brought writers from three teams along through working sessions rather than a top-down mandate.
- **Result:** The task-based IA shipped across the doc set, and the disambiguation pattern I designed there is one I still reach for when products overlap in confusing ways.