Instantiation of generate-exam.md for the standard final. Read that file for the LaTeX conventions and workflow; use these parameters.

Parameter	Value
Exam type	final
Agenda	docs/agenda/<term>.md, Meetings 9 onward (supervised models, deep learning and nPrint, unsupervised learning, generative models), with light reference back to the pipeline material. The final is not cumulative: the first half of the course (data acquisition, representation, preparation, the pipeline) belongs to the midterm, and the instructor’s “cumulative only as far as life is cumulative” line in class is a joke, not a scope statement
Past exams	docs/final/<YYYY>/ — the two or three most recent. From 2026 on, each also carries rubric.yaml (how it was scored), grading-notes.md (decisions made while giving it), and analysis.md (where the class split from the key): read those three first, and do not reuse a question they mark as ambiguous
Length	4 pages, 50 points, designed for 30–40 minutes; ends with a 2-point feedback section
Must include	questions on the hands-ons the agenda records in the meeting range (typically regression, trees and ensembles, deep learning, PCA, clustering, generative models) and on the project deliverables; every final flag in the range
Instructor output	/Users/feamster/Dropbox/Tmp/exams/ml-systems/final/<YYYY>/ (not committed until after the exam; when the instructor publishes it to docs/final/<YYYY>/, rubric.yaml, grading-notes.md, and analysis.md from the grading directory in instructor-data go with it, never the scores or review pages)

For the three-week offering, which has a single combined exam, use generate-combined.md instead.