Work with data using DataPressr

DataPressr provides AI assistant skills for turning a data question or raw source into a reproducible dataset, then a data story. Start with the setup guide, or pick the step you need below.

From a question to a story

┌───────────────────────┐
│ Find a source         │   datapressr: capture + discovery guide
└───────────┬───────────┘
            ↓
┌───────────────────────┐
│ Build a clean dataset │   datapressr: archive + structure + validate
└───────────┬───────────┘
            ↓
┌───────────────────────┐
│ Explore the findings  │   datapressr: enrich
└───────────┬───────────┘
            ↓
┌───────────────────────┐
│ Tell a data story     │   datapressr: story
└───────────────────────┘

The labels name the DataPressr skills (plus the discovery guide) that help your AI assistant: preserving sources, making data reproducible, checking the package, exploring findings and turning an argument into charts and prose. Start with your question or idea; you review the source, the numbers and the story's claims.

You can stop at a useful dataset. Use init to scaffold it and push to publish it to DataHub when ready; publication is optional and does not require a story. If you already have clean data, start at Explore the findings. The source-discovery guide helps you find evidence before the extraction skills begin.

Get started

1. Set up your AI assistant — install the skills, create a dataset, validate it and publish to DataHub.
2. Understand the dataset lifecycle — move from a saved idea to archived, structured and enriched data, with a clear quality bar at each stage.
3. Choose the right structure — distinguish catalogs, datasets and data files. For a source containing many datasets, use the catalog-as-repository pattern.

Find and prepare a source

Use the source-discovery playbook to turn a question into an extraction plan: compare sources, check licensing, establish coverage and record gaps before building.

Two worked examples show how that works in practice:

• Tesla quarterly deliveries — find releases through SEC EDGAR and adapt the question to what the source actually reports.
• U.S. natural hazard statistics — recover tables from PDFs when there is no machine-readable index and the source's own listing contains mistakes.

Once you have a source, use the skill playbooks for archive, structure and enrich. They cover raw snapshots, tidy CSVs, typed metadata and initial analysis.

Make a data story

• Story craft — choose one argument, support it with checked numbers and charts, and explain its limits.
• Charting — build reproducible story charts with Observable Plot and static SVG; use dataset views for initial exploration.
• Voice guide — write plainly, lead with the finding and let the numbers carry it.

Browse the datasets and finished stories for examples of the results.