docs: use case playbooks (persona-driven evaluation guides) #96

Summary

Create persona-driven use case playbooks — short, practical guides that answer specific questions a developer has when they sit down to evaluate their agent or model.

Motivation

Existing tutorials are organized by evaluation scenario (model-direct, RAG, agent workflow). But developers think in terms of their situation: "I just built a Foundry agent, how do I test it?" or "I need to compare two models before choosing one." Playbooks bridge this gap by starting from the developer's context and pointing them to the right bundle, dataset shape, and run config.

Proposed Playbooks

Each playbook should be 1-2 pages, actionable, and link to the detailed tutorial:

1. "I just created a Foundry agent — how do I evaluate it?"

   • Foundry agent + conversational or model quality bundle
   • Minimal dataset, first eval run, interpret results

2. "My agent uses RAG — how do I verify groundedness?"

   • RAG bundle + dataset with context field
   • Groundedness, relevance, retrieval evaluators

3. "I want to compare GPT-4o vs GPT-4o-mini for my use case"

   • Model comparison workflow with agentops eval compare
   • Same dataset, two run configs, side-by-side report

4. "I need to ensure content safety before deploying"

   • Content safety bundle (violence, sexual, self-harm, hate)
   • Adversarial dataset patterns, threshold recommendations

5. "I have an HTTP agent (LangGraph / LangChain / ACA)"

   • HTTP backend setup, request_field / response_field mapping
   • Tool calls extraction for agent-with-tools scenarios

6. "I want to gate PRs on evaluation quality"

   • agentops workflow generate, exit code contract
   • Threshold strategy, AZURE_AI_FOUNDRY_PROJECT_ENDPOINT as secret

Format

Each playbook follows the same structure:

• Situation — 1-sentence description of the developer's context
• What you need — prerequisites (agent deployed, model available, etc.)
• Steps — numbered, with exact CLI commands
• Expected output — what results.json and report.md will show
• Next steps — links to detailed tutorials and related playbooks

Acceptance Criteria

• At least 6 playbooks covering the scenarios above
• Each playbook is self-contained and actionable
• Consistent format across all playbooks
• Links to detailed tutorials for deeper reading

Context

• These playbooks complement the E2E adoption guide (docs: end-to-end adoption guide (workshop-ready, from zero to production) #95) — the guide is the sequential path, playbooks are the quick-reference cards
• Existing tutorials provide the deep-dive content that playbooks link to