CourseLarge Language Models: The Complete Course · Module 10: Evaluation · part 70 of 103
Part 70 · Module 10: Evaluation

Lesson 2: Noise, sample size and fair comparisons

22 min read·13 Oct 2026
Module 10 · Lesson 2 · About 50 minutes · Needs: Module 1 Lesson 4 (the Wilson interval, the noise floor), Module 4 Lessons 2 and 4 (the control prompt, the boundary rule, discordant tickets)

The situation

It is Monday 3 August 2026, week 10, and shadow mode starts today. Two triage versions compete: v1, Module 4's control prompt, and v2, the same prompt plus the boundary rule written for T-1004, Yuki Sato's refund request 20 days into an annual plan.

Your rehearsal on the local instrument model gives v1 12 of 24 and v2 9. Maya asks: "Is v2 worse, or is that luck?" This lesson builds the answer: an interval on every score, the noise floor, the number of tickets a difference needs, error analysis, and a paired comparison. T-1004 is the running example.

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.