CourseLarge Language Models: The Complete Course · Module 4: Prompt Engineering · part 32 of 103
Part 32 · Module 4: Prompt Engineering

Lab: Prompt files, a frozen split and a harness

39 min read·13 Oct 2026
Module 4 · Lab · About 60 minutes · Needs: lessons 1 to 4 of this module, and the Module 1 and Module 2 labs (or the cells in Step 1)

What you are building

THE MODULE 4 PIECE OF THE REPOSITORY prompts/ three files prompts.py load, render the model via ask() harness.py parse, score split_manifest.json frozen 8 June runs/results.jsonl score + ids
Prompt files and code (blue) send T-1001 to the model (amber); the harness scores replies against the frozen split (grey) and logs every result with its ids (green).

It is Friday 26 June 2026, week 4. This week the triage prompt became a file (Lesson 1), the 36 tickets were frozen with a control prompt beside every run (Lesson 2), techniques were measured on the dev set (Lesson 3), and every change was made one at a time and logged (Lesson 4). All of it lived in notebook cells. This lab moves it into the repository: prompts/ holds the prompt files, supportdesk/prompts.py loads and renders them, supportdesk/harness.py freezes the split, runs a prompt file over it, scores it and logs the result, and runs/ keeps the evidence. The running example is T-1001, Ana Duarte's duplicate charge. Forty-nine calls go to the hosted model, all in Step 6, all through the course's guard.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.