Lesson 1 — Quality, Evidence, and Responsible AI

Duration: 120 minutes

Included delivery track: Follow LAB_SETUP.md for the supplied files and local simulation. Required activities use these materials; connected deployments are optional extensions. Instructor debrief: answer key. Real OAuth, cloud releases and payments are simulated or discussed through evidence packets. Audience: first-time and experienced learners together
Deck: 01-quality-and-evidence
Lab: The AI Claim Detective

Learners will be able to

distinguish a request, observation, hypothesis, finding, and verdict;
write a defect report another person can reproduce;
use an AI assistant without treating its answer as proof; and
name data/actions that are off limits in a training environment.

Instructor preparation

Use fixtures/claim-packet.md and the evidence-packet template. E01–E05 support three claim classifications. Capture an analogous local screenshot separately; no database record or console export is supplied. Prepare ten short fact/inference/question statements for guided practice.

Agenda

Time	Facilitation
0–10	Arrival: show a beautiful but wrong AI answer. Ask, “Would you ship on this?” Learners vote and explain why.
10–25	Teach the ladder: observation → hypothesis → reproducible finding → evidence-backed recommendation. Emphasize that confidence is not evidence.
25–40	Live demo: ask an approved assistant to summarize a small bug packet; compare its claims to the sources. Model the phrase “I cannot prove that from the supplied evidence.”
40–50	Pairs classify ten statements as fact, inference, question, or unsupported claim. Review together.
50–60	Break. Instructor privately checks that every learner can explain why screenshots alone may mislead.
60–95	Run the lab in pairs. Driver reads sources; verifier maintains the evidence table. Switch at midpoint.
95–110	Gallery walk: teams review another team’s packet and circle one claim that lacks proof.
110–115	Debrief: what AI behavior was useful? What behavior would be dangerous?
115–120	Exit ticket: one sentence distinguishing a lead from a defect, plus one safety rule.

Teaching notes

Keep the first session human-centered: good QA protects a customer from a bad experience and protects a team from bad decisions. Do not introduce prompt tricks as the core skill. The core skill is calibrated judgment.

If students finish early, ask them to improve an overconfident AI answer using only citations to the supplied sources. If students need support, give them the sentence starters: “I observed…”, “I expected… because…”, and “This evidence does not establish…”.

Assessment

A learner is ready to continue when their packet includes an expected outcome, observed outcome, reproducible steps, and a clear limit on what the evidence does not prove.