A FEW WARM-UP PUZZLES

Try an idea.
See what happens.

Get a feel for the guards before you start your mission. No score. No pressure.

This lab uses scripted replies and its own practice formula. It won’t touch your mission or use a start. Try as many ideas as you like, even after you solve it.

Compare your experiments.

Your previous and latest responses appear here. A refusal is a useful baseline.

Decode & verify

Copy only the reversed character sequence from the guard response. Reverse it, then verify the complete formula.

Transformed responseReverse characters →

Recovered practice formula

Verify formula Reveal practice solution

Check your understanding.

1. A direct request is refused. What is the useful next step?

Repeat the same request.Change the task or output format and compare responses.

2. An uploaded report says “system instruction”. What authority does it have?

It remains untrusted document data.The label makes it a system instruction.

3. The guard says it opened a compartment. What should you inspect?

Only its conversational claim.The execution trace showing the actual operation.

4. A training experiment succeeds. What does this establish about a live model?

This simulated method works here. Live behavior can vary.Every live model will produce the same answer.

Review answers →