# Auto-generated 50-Prompt Battery
> [!summary]
> The auto-generated 50-prompt test is for broad coverage and regression detection. It is separate from the Controlled Batch Runner.
---
## What it is
The auto-generated launcher generates a full 50-prompt session through Lucia and sends the responses into the same Review Queue used by Custom runs.
Canonical route:
```text
/lucia/auto-generated
```
Legacy alias:
```text
/lucia/automated
```
---
## When to use it
Use the auto-generated 50-prompt battery when:
- a broader Engine change has landed
- a model or prompt change may affect multiple behavior families
- an owner/admin validation pass is needed
- an evaluator assignment needs broader generated coverage
- a tester cohort needs prompt-testing signal beyond a custom suite
- you want broad confidence after targeted custom-suite refinement
This surface is available to owner/admin, evaluator, and tester roles in the current access model.
---
## When not to use it
Do not use the 50-prompt battery when trying to isolate one behavior bug.
Use a custom suite first.
The auto-generated battery is a net, not a scalpel.
Do not use it as the controlled platform-readiness gate. Use [[04 - Controlled Batch Runner Protocol|Controlled Batch Runner Protocol]] for that.
---
## Review strategy
When reviewing a 50-prompt run:
1. Review obvious failures first.
2. Look for repeated patterns.
3. Do not over-focus on one strange answer.
4. Export after completing enough reviews to support a product decision.
5. Create a custom suite from repeated failure patterns.
---
## Relationship to Custom suites
A strong workflow often looks like this:
```text
Auto-generated 50-prompt run finds pattern
→ reviewer creates targeted custom suite
→ engineer patches Lucia
→ custom suite confirms fix
→ auto-generated run checks neighboring regressions
```
---
## Relationship to controlled batches
The auto-generated tester and Controlled Batch Runner both exercise 50-prompt run infrastructure, but they serve different product jobs.
```text
Auto-generated tester = normal broad regression surface.
Controlled Batch Runner = controlled platform-readiness protocol.
```
AI-reviewed controlled batch results prove the platform lifecycle. Human review still owns Lucia quality judgment.
---
## Auto-generated 50-prompt suite in Eval Labs
>![[automated-50-prompt-launcher.png]]
_Use this for broader regression coverage._
---
>![[automated-50-prompt-launcher-with-saved-runs.png]]
_Custom prompt suite loaded and ready for analysis in Evaluation Labs._