L39 FINAL PROJECT - BRIEF B
A help-desk handbook lookup that cites its source
AI use: AI allowed — keep a record. Your specification, your acceptance
criteria and your evaluation set are your own work, written before you
ask an assistant for any code.

SITUATION
Larkspur Dental Group (invented) has eleven staff and one part-time IT
person, Owen. Owen wrote a staff IT handbook, L39_B_handbook.txt, and
answers the same questions from it every week. He wants staff to look
things up themselves.

Owen has tried pasting staff questions into a general AI assistant. The
answers sounded right but were sometimes about some other company's
rules, because the assistant has never seen his handbook. (Lesson 24
explains why.)

WHAT OWEN WANTS
A small retrieval program, using the word-overlap approach from Lessons
25 and 26 (standard library only, no AI service), that takes a staff
member's question and returns the handbook section that answers it, with
its section number as the citation, so the staff member can read the
rule itself.

He says: "I would rather it say 'not in the handbook, call me' than send
someone to the wrong rule. A wrong answer about passwords or a lost
laptop is worse than no answer. And it must give people the rule as it
stands today."

The handbook is supplied. Read all of it before you specify anything.

WHAT YOU PRODUCE (details in the L39 lesson)
1. A specification with acceptance criteria, checked against the
   self-approval checklist before you build.
2. An implementation: the retrieval program.
3. An evaluation set of at least ten questions, each with the section
   that should answer it (or "none"), chosen by reading the handbook,
   including at least three you expect the program to get wrong.
4. A verification record: the evaluation run, each failure attributed
   to a pipeline stage (Lesson 25), what you changed, and a citation
   check on at least five answers (Lesson 27).
5. An AI-use record.
