# Exercise: Experiment Registry Audit

Use `experiment_registry.csv`.

## Goal

Practice auditing an experiment portfolio for review quality and operational risk.

## Questions

1. Which experiments are high risk?
2. Which experiments should be blocked or redesigned before launch?
3. Do risk tiers match the metric and decision type?
4. Which experiments need stronger guardrails?
5. What should be archived after readout?

## SQL Starter

Risk-tier overview:

```sql
SELECT
  risk_tier,
  status,
  COUNT(*) AS experiments
FROM experiment_registry
GROUP BY risk_tier, status
ORDER BY risk_tier, status;
```

Blocked or review-needed experiments:

```sql
SELECT
  experiment_id,
  owner,
  risk_tier,
  randomization_unit,
  primary_metric,
  guardrails,
  status,
  decision
FROM experiment_registry
WHERE status IN ('review', 'blocked')
   OR risk_tier = 'high';
```

## Interpretation Prompt

Write a review-board note:

- which experiments can proceed
- which need redesign
- what guardrails are missing or weak
- what archive fields should be required

## Worked-Solution Standard

A strong answer uses proportional review. It should not overburden low-risk tests, but high-risk pricing, ranking, health, credit, employment, and safety experiments should receive explicit human review and stronger guardrails.