Pipeline
A release moves through source inventory, rights review, immutable raw storage, field-level normalization, geography validation, aggregation and release review.
| Stage | Required record |
|---|---|
| Source | Owner, URL, terms, coverage and version |
| Raw | Original ID, source version and unchanged fields |
| Normalize | Documented mappings with nulls preserved |
| Validate | Counts, coordinates, duplicates and city boundary |
| Release | Aggregate version, caveats and downloadable files |
Counting rules
A crash count is distinct from a person injured or killed. Severity groups follow source definitions. A pedestrian, bicycle, motorcycle or commercial-vehicle flag appears only when a documented field supports it.
- Use a stable source collision identifier
- Do not convert missing coordinates into estimated points
- Keep city and county filters reproducible
- Suppress incomplete dimensions rather than guess
- Compare release totals with source summaries
Known limitations
Police-reported datasets can omit unreported crashes, recent records may be incomplete, and geocoded locations can differ from officer-reported coordinates. Each release must state which limitations apply.