# Exercise: Onboarding Experiment Readout

Use `onboarding_experiment.csv`.

## Goal

Practice turning an experiment dataset into a decision-oriented readout.

## Questions

1. What is activation within 7 days by assigned arm?
2. What is the absolute treatment-control difference?
3. Do support tickets look worse in treatment?
4. Does trial-to-paid conversion move in the same direction as activation?
5. Are there acquisition channels where the effect looks meaningfully different?

## SQL Starter

```sql
SELECT
  assigned_arm,
  COUNT(*) AS workspaces,
  AVG(activated_7d) AS activation_rate,
  AVG(support_tickets) AS support_tickets_per_workspace,
  AVG(trial_to_paid_30d) AS paid_conversion_rate
FROM onboarding_experiment
GROUP BY assigned_arm;
```

Channel check:

```sql
SELECT
  acquisition_channel,
  assigned_arm,
  COUNT(*) AS workspaces,
  AVG(activated_7d) AS activation_rate
FROM onboarding_experiment
GROUP BY acquisition_channel, assigned_arm
ORDER BY acquisition_channel, assigned_arm;
```

## Interpretation Prompt

Write a readout that starts with validity and guardrails before making a ship/no-ship recommendation.

Use this structure:

- validity caveats
- primary metric result
- guardrail result
- segment caveat
- decision

## Worked-Solution Standard

A strong answer does not overclaim from a tiny dataset. It should say what the practice data suggests while noting that a real launch would require pre-specified power, SRM checks, and complete exposure logging.