Methods

How we work

1. Start from a primary government file, not from another newsroom’s spreadsheet.
2. Record the URL, fetch time, HTTP status, and a hash of the raw bytes.
3. Recompute every published number from those rows at build time.
4. If a file is missing, show an empty state. Do not interpolate.
5. Keep geometry out of the database except points. Polygons live as simplified snapshots.
6. Arrest records: drop names, dates of birth, and physical descriptions before anything is stored.
7. Flag series seams (UCR SRS → NIBRS in 2025; CCRS agency mix) next to the chart.
8. Correct in public on the corrections page.

Limits that will not go away on their own

• No BPD incident extract — needs a CPRA request.
• No structured council votes — needs minutes PDF parsing.
• No campaign-finance bulk file — the city eFile API is undocumented; we will not reverse-engineer it.
• Census ACS needs an API key for automated pulls.

HTTP API notes