CHAPTER 45: Expose retrieval and evidence boundaries

Boards: 45a and 45b.
Read the surrounding chapter before this lab. All examples are original and synthetic.

PREDICT
Predict the overlap scores for desk delivery. Then read all three knowledge files. The printed policy lets you compare source text with the deliberately false tomorrow promise.

RUN
From the companion root, run:
python visual_labs/ch45/walkthrough.py
Use python3 if that is your configured command.

COMPARE
Compare the printed values and explanations with both boards. Read expected_output.txt after making your prediction. An assert checks a known expected result and stops if it differs. Temporary output folders are removed after the demonstrations; original course files are not edited.

CHANGE ONE THING
Change the query to elephants: expect no sources. Change the false draft citation to invented_policy: expect Unknown source ID. Explain why the original false draft passed mechanical validation despite being unacceptable to a reviewer.

EXPLAIN
State which input changed, which instruction handled that change, and why the resulting output is appropriate. If your prediction differs, identify the first step at which it diverges. Return to the surrounding lesson before adding more complexity.
