Unit 4 project: Audit a tool for fairness

You counted Ambri leaves, drew kangris, and argued about who answers for a wrong tick. Today your group audits a real tool, the way a fairness inspector would: find out where its data came from, test it on the world you live in, count the results, and say what should change.

Work in groups of 3–4. One period. Everything in fairness-audit.ods.

Pick one your teacher approves — no accounts, nothing about people:

• Quick, Draw! — does it know our objects?
• Your own Teachable Machine model from Unit 2 — which classes does it fail?
• A phone keyboard's suggestions (on the teacher's phone) — does it know our words and names?
• A speech or translation tool, run by the teacher — does it know our place names and accents?

1. Dataset card — the nine questions from Lesson 4.1. Write not stated wherever the tool doesn't say.
2. Choose the tests on purpose. At least ten cases in two groups: five that the tool's makers probably had plenty of examples of, five that are ours — kangri, samovar, Kashmiri place names, a local handwriting style, dim winter light.
3. Predict each result first. Then run every test once and record case · group · predicted · what the tool did · right?
4. Count by group: how many right in each? That's your evidence.
5. Accountability note — three lines: the bias check that should have been done, where a person should review it, and what it should tell users plainly.
6. One rule: "Before this tool is used in our school, it must ___."

Each person writes two sentences to the tool's builders: one thing it did well, and the one group it should learn from next, with your evidence. Sign with your roll number. This part is marked for you alone.

Each group reads out its two counts — "ours" vs "theirs" — and its rule.