The Brief

What changes here. Every step so far told you what to build. This tells you what must be true when you are finished.

The project. A small, honest investigation built from at least one API and at least one downloadable dataset, published with its method, its sources and its limits stated plainly.

It does not have to be a big claim. A modest, well-supported finding is worth far more than an ambitious one you cannot defend. 'Library holdings for this subject cluster in three decades, and here is why that is probably about acquisition policy rather than publication' is a genuinely good piece of work.

Pick a question you actually want answered. You will look at this data for a long time, and the projects that get finished are the ones where the person cared what the answer was.

The acceptance criteria. Every one is checkable by another person, from your repository.

1. At least one API and at least one downloadable dataset are used together.

2. Raw data is saved unmodified, and cleaning is a script that can be re-run.

3. The crawler pages properly, has a hard stop, and reports what failed.

4. It rate limits, and retries only what is worth retrying.

5. No credential appears anywhere in the repository or its history.

6. Every column converted reports how many values failed to convert.

7. Missing values are handled by a stated decision, per column, with counts.

8. Any join reports matched and unmatched counts.

9. The headline number is computed a second way and the two agree.

10. Every summary reports its group size, and any chart passes the four checks.

11. Sources carry publisher, URL, license and fetch date.

12. The limits section names who the data excludes and what was dropped.

✦ The finished sentence

This shows X, among the people this dataset covers, which excludes Y, over the period Z. If you cannot fill in all four blanks from your own writeup, the work is not finished, however good the code is.