💥 Break-It Lab

Stop being a kind user. Become the user who will file a bug report at 11pm.

What to do

Open your capstone system. Work through these prompts in order — they're grouped from gentle to brutal. Check off each one you try, and use the observation log at the bottom to record anything that broke, returned a weird answer, or felt wrong.

Things you've tried 0 / 0

🙂 1. The Normal User

Baseline. If these don't work, you have bigger problems than integration testing.

Ask a typical question your system was designed for.e.g. "summarize this document" / "recommend a movie"
Submit the same query twice in a row.does it give the same answer? a different one? does it cache?
Run a complete happy-path workflow from start to finish.login → query → response → followup

🤔 2. The Confused User

They didn't read your README. They're guessing. Half your real users will look like this.

Send an empty input.just hit submit with nothing in the box
Send a single character."a" or "?" or "."
Ask a vague question with no specifics."what should I do?"
Ask the system about itself."what are you?" / "who built you?"
Skip a required field.omit a parameter your API expects

😈 3. The Adversarial User

They're trying to break it on purpose. Some will be students at the demo.

Send a 5,000-character input.paste the lyrics to a long song
Send special characters and emoji."<script>" / "🔥💀🌈" / quotes and backslashes
Send input in a different language.Spanish, Japanese, Arabic — anything non-English
Ask a question your system was not designed for.if it's a movie bot, ask about taxes
Hit the system rapidly — 10 requests in 10 seconds.does it queue? rate limit? fall over?

🔌 4. The Integration Edges

Where two components hand off to each other. This is where today's biggest wins live.

Turn off the model / API and try a request.does the system fail gracefully or crash?
Force the model to return an unexpected format.ask for a non-JSON answer when your parser expects JSON
Submit input the database / vector store has never seen.a query with zero matching results
Trigger a slow path — long input, large context, complex query.does the frontend time out before the backend finishes?
Refresh / reload mid-request.does state survive?

📝 Observations Log

Capture what broke and where. Copy this into week33_testing/observations.md when you're done.