GentlePrince-ng/nhw-data-studio

Upload a spreadsheet, get the read on it: automatic column typing, ranked findings, chosen charts, Nigeria state/LGA maps, and a self-contained HTML report. Built for Kobo and DHIS2 exports. # NHW Data Studio Upload a spreadsheet, get the read on it, take a report away. No code, no setup per dataset, no waiting for an analyst to be free. ```bash pip install -r requirements.txt streamlit run app/studio_app.py ``` Then drag any `.xlsx` / `.xlsm` / `.csv` / `.tsv` into the sidebar. The Nigeria state and LGA maps work out of the box — boundaries are included under CC BY 4.0, see The maps. To see it work before pointing it at real data: ```bash python samples/make_sample.py # a 10,180-row Kobo-style survey with planted defects python tests/test_pipeline.py # read -> type -> analyse -> chart -> export, asserted python tests/test_edges.py # every awkward upload we have met, + 100,000 rows ``` Both exit non-zero on failure, so `python tests/test_edges.py && echo ok` is the check — not reading the output, which is how two red suites once got reported as green. On the sample that runs in about **six seconds** end to end — including drawing both maps and building the downloadable report. The largest real workbooks in the archive — 66,000 rows, or 576 columns across 21 sheets — take under ten. ## What it does **Types every column** — dates, categories, numbers, binaries, identifiers, free text, geography, empty, constant — and shows you its decisions on the Data health page. Nothing about the analysis is hidden. **Builds metrics, not a profile.** Columns are split into **measures** (things you aggregate — visits, wait time, doses) and **dimensions** (things you slice by — state, facility type, month). The top row carries those measures with period-over-period change and a sparkline, not the record count and the percentage of cells filled. Whether a column is totalled or averaged is a guess from its name, it is printed on every figure, and you can change it on the page. **Answers "what moved", not just "what is biggest".** Each measure gets three views: the level over time, the split by dimension, and a diverging bar of the *change* per di …