fix(tutorial): lowercase both operands in the Exercise 3.1 grader

[shoemoney]

Summary

The Exercise 3.1 grader in Chapter 3 lowercases one operand and not the other, so a correct answer is graded as a failure.

def grade_exercise(text):
    if "incorrect" in text or "not correct" in text.lower():

The exercise asks the learner to get Claude to judge this as incorrectly solved:

2x - 3 = 9
2x = 6
x = 3

A response that opens with a capitalised "Incorrect" — the most natural way to begin that sentence, and what a role-prompted math teacher tends to produce — misses the first operand entirely, and "Incorrect." does not contain "not correct" either:

grade_exercise("Incorrect. 2x - 3 = 9 means 2x = 12, so x = 6.")  # -> False
grade_exercise("No, the solution is INCORRECT.")                  # -> False
grade_exercise("Incorrect!")                                      # -> False
grade_exercise("incorrect - the second step is wrong.")           # -> True

Three of those four are correct answers marked wrong. A learner who solves the exercise as intended sees This exercise has been correctly solved: False and reasonably concludes their prompt is the problem.

The change

One character per file: text → text.lower() on the first operand.

-    if "incorrect" in text or "not correct" in text.lower():
+    if "incorrect" in text.lower() or "not correct" in text.lower():

This matches what the second operand in the same expression already does, and the convention everywhere else in the tutorial — "hola" in text.lower() in Chapter 2, re.search("brown", text.lower()) in Chapter 4. The graders that don't lowercase are matching deliberately capitalised strings ("Warrior", "49-fold"), which is not the case here.

Applied to all three copies of the chapter, which carry the identical line:

• prompt_engineering_interactive_tutorial/Anthropic 1P/03_Assigning_Roles_Role_Prompting.ipynb
• prompt_engineering_interactive_tutorial/AmazonBedrock/anthropic/03_Assigning_Roles_Role_Prompting.ipynb
• prompt_engineering_interactive_tutorial/AmazonBedrock/boto3/03_Assigning_Roles_Role_Prompting.ipynb

Testing

Ran the grader verbatim before and after. After the change all five phrasings of a correct answer pass, and responses that judge the equation as correctly solved still return False — "solved correctly" contains neither incorrect nor not correct, so no false positives are introduced.

Only the one line changed in each notebook: git diff --numstat is 1 1 per file, with no cell-output or metadata churn.

Not in this PR

real_world_prompting/02_medical_prompt.ipynb has an unrelated defect — extract_action_items passes None into json.loads when the <summary> tags are missing, and the resulting TypeError is not caught by the surrounding except json.JSONDecodeError, so it aborts the batch loop. Separate concern, happy to send it separately if useful.