Data Track — Week 2 Assignment (Template)

The HackYourFuture Data Track Week 2 assignment: Refactoring to a Clean Pipeline.

👩‍🎓 Students: you are in the wrong place. Do not fork or use this template. Go to your cohort's assignment repo under HackYourAssignment (e.g. c55-data-week2, c56-data-week2, …). Your teacher posts the exact link in your cohort channel. Fork the cohort repo, branch, and open a PR back to it. Full instructions live in the Week 2 Assignment on Notion.

For instructors / track maintainers

This repo is the upstream template for the Week 2 assignment. At the start of each cohort, generate a cohort-specific repo under the HackYourAssignment org from this template (GitHub: Use this template → Create a new repository, owner = HackYourAssignment, name = c<NN>-data-week2). Students then fork that cohort repo and open PRs back to it; the auto-grader runs on every push.

Edits to the assignment, dataset, or grader belong here on the template, not on the cohort copies.

Tasks at a glance

Task	Folder	Points	What you build
Task 1 — Cleaner Pipeline	task-1/	60	A modular Python pipeline with config.py (env-var loading), models.py (Transaction dataclass with __post_init__ validation), transforms.py (4+ pure composable functions, no mutation), pipeline.py (orchestrator), and tests/test_transforms.py (4+ pytest tests). Reads data/messy_sales.csv, writes output/clean_sales.csv.
Task 2 — AI Debug Report	task-2/	20	Document one debugging session where you used an LLM to fix a bug. Fill in the four sections of AI_DEBUG.md.
Task 3 — HYF Azure proof	task-3/	20	Confirm your HYF Azure tenant access still works. Screenshot proof at task-3/azure_proof.png (or .jpg / .jpeg) showing resource group + region + €0 cost.

Total: 100 · Passing: 60.

Repository layout

.
├── task-1/
│   ├── data/
│   │   └── messy_sales.csv      # the dataset (committed; do not edit)
│   ├── src/
│   │   ├── config.py            # env-var loader — fill in TODOs
│   │   ├── models.py            # Transaction dataclass — fill in TODOs
│   │   ├── transforms.py        # 4 pure transform functions — fill in TODOs
│   │   └── pipeline.py          # orchestrator — fill in TODOs
│   ├── tests/
│   │   └── test_transforms.py   # 4 pytest tests — fill in TODOs
│   ├── output/                  # your pipeline writes clean_sales.csv here (gitignored)
│   ├── .env.example             # copy to .env (gitignored) before running
│   └── requirements.txt         # python3 -m pip install -r requirements.txt
├── task-2/
│   └── AI_DEBUG.md              # fill in the four sections
├── task-3/
│   └── azure_proof.png          # add your screenshot here
├── .hyf/
│   └── test.sh                  # auto-grader (read it to see exactly what it checks)
└── .github/workflows/
    └── grade-assignment.yml     # runs .hyf/test.sh on every PR

Run the grader locally

Before opening a PR, run the same checks the auto-grader runs:

cd task-1
python3 -m pip install -r requirements.txt
cp .env.example .env
cd ..
bash .hyf/test.sh
cat .hyf/score.json

The grader prints a per-task breakdown so you can see exactly which check failed and why. The PR-time grader does the same — your local run and the CI run are identical.

Scoring ladder (Task 1)

The grader awards points incrementally so partial credit is meaningful:

• 10/60 — required files exist (config.py, models.py, transforms.py, pipeline.py, tests/test_transforms.py, .env.example).
• 20/60 — python -m src.pipeline runs from task-1/ without crashing (the grader injects INPUT_PATH and OUTPUT_PATH inline; your local .env is not used during grading).
• 40/60 — output/clean_sales.csv passes structural checks: 12 rows (15 input − 3 invalid/zero-quantity), lowercased emails, title-cased product names, "Unknown" filled in for missing categories, revenue and vat columns present and correctly calculated.
• 60/60 — code looks engineered: models.py defines a @dataclass with __post_init__; transforms.py uses the {**row, ...} spread pattern (no mutation); pytest tests/ reports all tests passing.

The 40-point cap exists to stop a 5-line script that hardcodes the expected JSON from getting full marks. Real engineering patterns (dataclass + spread + tests) are required for the top 20 points.