The 42nd Period | STEM Education Hub

🤝 Your Client

Real client? If your teacher arranged one, use what they tell you and skip the practice card below. The deliverable and rubric are the same. Practice client? Enter your team code. Your teacher may give you one, or you can make one up. The same code always gives the same client.

Team code

Loading…

Working with a real client? Questions to ask them

Your teacher arranges the client, and a guardian or school staff member knows about every meeting. Never contact a stranger you only know online.
What does the organization do, who works there, and what does a normal week look like?
Which tasks take the most time? What happens if each one goes wrong, and who looks at it today?
Is a tool already in use, or being considered? What does the vendor promise? What have they actually seen it do?
What information is private (customers, students, families, children), and what does the client's own rule say about it?
Ask permission to use what they tell you in class. Leave out names and anything confidential. Your teacher reads your guide before anything is shared with the client.

🧰Tools You Already Have

from the rest of the unit

From	What you can use
3.2.1 Substitute or complement?	The four questions (Routine or judgment, How costly is a mistake, Does it depend on trust, Who checks the AI's work) give a level: 1 AI does it, 2 AI drafts, a person reviews each, 3 AI assists, a person decides, 4 A person does it. Two stop signs: a hard-to-undo mistake means level 2 at least, and no one checking means level 3 at least.
3.2.2 Talent stack	Four strands (Make, Know, Connect, Steer), and the difference between a claim and evidence.
3.2.3 Task X-ray prep	The camera test for tasks, interview questions that get true answers, and the safe-interview rules.
3.2.4 Task X-ray report	Reach and gap: can versus should. Name your evidence, mark guesses as guesses, and say what you don't know.
3.1.2 Data footprint	Ask what a tool collects, who else sees it, and what should never go in.

Automate, assist, human: automate is level 1, assist is levels 2 and 3, and keep human is level 4.

🔒Constraints

Work in a team of three or four. Every member speaks in the presentation and signs the AI use log.
Use your client's facts. Where you assume something, say so. Where the client's tool is untested, say it's untested.
Every decision needs a reason, not just a level. “It's faster” isn't a reason by itself.
No private data in a chatbot. No real names, student or customer details, health information, contact lists or anything confidential, from a real client or from the practice card. Use a made-up example instead.
A district-approved chatbot may help you draft or question your work. The decisions and the reasoning have to be yours. Write every chatbot use in your AI use log, and check anything it tells you about the client against the client's facts.
Promise nothing about anyone's job or about what the tool will do. Say what you decided and why, and what you don't know.
Your guide is two pages or fewer, plus the decision table, the log, and the failure analysis.
Your teacher reads everything before it goes to a real client. A guide is advice, and it isn't legal, medical or safety certification.

📋Deliverable: What You Hand In

The AI-use guide. A decision table that sorts every task in the client's list: the level (1 to 4), the reason from the four questions and the client's facts, who checks, and how. Then three rules the staff can follow (for example, Always, Never, If unsure) and the data that never goes into the tool. Plain language, two pages or fewer.
The AI-use log. A one-page log the client keeps when they use the tool. It has columns that fit this client (at least: date, who, task, what the tool produced, what was checked and by whom, what was changed, and the outcome), three sample entries, and a line saying when to review it and what pattern would change a rule.
The failure-point analysis. At least three ways the tool could fail, including one that looks right but is wrong. For each: how it would be noticed, who notices and when, what happens next, who can turn the tool off, and how the client restarts safely.
The peer test. Another team runs your guide on three test cards and records where it helped and where they got stuck. You make at least two real changes and record what you changed and why. (Cards are in the Day 5 panel below.)
The presentation. Five minutes, then three minutes of questions. Cover your hardest decision, what the client should check in the first week, and what would make you change the guide. Every member speaks.

Your AI use log for any chatbot you used during the problem goes in with your work. It's a separate sheet from the client's log above.

🗓️Eight-Day Plan

a guide, not a script

Day 1	Meet your client. Read the card (or interview a real client). List what you know and what you'd need to ask.
Day 2	Sort every task with the four questions. Give each a level and a reason.
Day 3	Plan the checks: who looks, when, how. Decide the three rules and what data never goes in.
Day 4	Failure points: at least three, with one that looks right but is wrong. Add the fallback and the off switch.
Day 5	Draft the guide and the log. Run another team's guide on the test cards.
Day 6	Revise from the peer test. Read the surprise. Decide what it changes.
Day 7	Finish the guide and log. Practice the five-minute defense.
Day 8	Present and answer questions.

🧪Peer Test Cards

open on Day 5

On Day 5 you test another team's guide. Pick their client here and use only their guide to decide what to do with each card. Don't open this for your own client before then. It's more useful to be surprised. With a real client, each team writes three cards for the team that tests them: a wrong answer that looks right, a tool that stops, and a request the tool shouldn't handle.

Open the cards

The client you're testing for

📊Grading Rubric: 100 pts

A starting point. Adjust weights to fit your class before assigning.

Category	Weight	Exemplary	Proficient	Developing	Incomplete
Task Decisions	25 pts	Every task has a level (1 to 4) with a reason drawn from the four questions and the client's own facts. Stop signs are applied. At least one task is kept with a person and at least one is given to the tool, and each choice could be defended.	Every task has a level and a reason, but a reason is thin or doesn't use the client's facts, or a stop sign is missed.	Levels are given without reasons, or most reasons are “it's easier.”	Tasks aren't sorted.
Checking & Failure Plan	20 pts	Each check says who looks, when and how, and fits the client's time. At least three failure points, including one that looks right but is wrong, each with how it's noticed, what happens next, and who can switch the tool off and how to restart.	Checks and three failure points are present, but a check is vague (“someone will look”) or the quiet failure is missing.	One or two failure points, or no fallback.	No failure analysis.
AI-Use Guide & Log	15 pts	A busy staff member could follow the guide in plain language on two pages or fewer, with three rules and the data that never goes in. The log fits the client, has three sample entries, and says when to review it and what would change a rule.	Usable, with some jargon or a log that's generic.	Hard to follow, or the log is missing key columns.	No guide or no log.
Peer Test & Revision	10 pts	Another team ran the guide on three cards and recorded where it helped and where they were stuck. The team made at least two real changes and recorded what changed and why.	Test done and one real change recorded.	Test done, but changes are cosmetic or unrecorded.	No peer test.
Evidence & Honesty	15 pts	Uses the client's facts, marks assumptions and guesses as guesses, and states what the team doesn't know. Promises nothing about the future or about anyone's job. No private data is pasted into a chatbot, and every chatbot use is in the AI use log.	Mostly honest, with one assumption left unmarked or a claim that goes a little past the evidence.	Claims about the tool or the future that nothing supports.	Private data shared with a chatbot, or an AI use log is hidden or missing.
Presentation & Defense	15 pts	Clear in five minutes. Every member speaks. Answers questions with reasons, names the hardest decision, and says what would make the team change the guide.	Clear, but one member carries it or an answer has no reason.	Reads the guide aloud and struggles with questions.	Not presented.

📎Team Workbook

One workbook holds the decision table, the client's log, the failure-point sheet, the peer-test record and the presentation outline.

⬇ PDF 📄 Google Doc (make a copy)

← Back to Problem 3.3.1 🗺️ Back to Unit 3 Overview

📝 Problem 3.3.1 · Unit 3 of 3 · 11 Weeks