Skip to main content
Thirdwatchthirdwatch
Business & local data

Build a Used-Car Safety Dataset

Enrich used-car catalogs with model-level NHTSA recall and complaint context while keeping VIN decisions separate.

Jul 21, 2026 · 2 min read · 309 words
See the scraper →

A marketplace may hold thousands of listings but only hundreds of distinct vehicle configurations. That makes model-level safety enrichment practical: collect once per make, model, and year, then attach the context to every matching listing. The NHTSA Vehicle Safety Scraper supplies official recall campaigns and complaint signals without collecting VINs.

Build the enrichment table

Start from a normalized catalog dimension with one row per distinct configuration. Send those values in batches and request the record types the product actually needs. Store recalls and complaints in separate tables. Campaign number is the recall key; complaint number is the complaint key. A bridge table can connect each record to the catalog configuration that produced it.

For recalls, retain component, consequence, remedy, manufacturer, affected units, and report dates. For complaints, retain component, narrative, event dates, and severe indicators. Add collected_at to every snapshot and refresh on a schedule. A weekly run is a sensible default for active inventory, with slower refreshes for archived research datasets.

The presentation layer needs careful wording. “Recall campaigns associated with this model year” is defensible. “This car has an open recall” is not, because repair status and production-range applicability are vehicle-specific. Provide a prominent link to the official VIN lookup and advise shoppers to confirm with NHTSA or the manufacturer. Never infer a listing’s repair status from model-level campaign data.

Quality checks should flag unmatched model names, unusually empty results, duplicate campaigns, and large changes between snapshots. Complaint totals should not become a simplistic safety score; popular and older models have different exposure and reporting volume. Use them as research context with a clear disclaimer.

This architecture keeps collection efficient and separates public model-level evidence from sensitive or operational vehicle identifiers. It gives catalog teams a maintainable safety layer while preserving the boundary between useful enrichment and a definitive assessment of one used vehicle.

Frequently asked questions

Should model-level results be shown as an unresolved recall?

No. Campaign context is model-level; unresolved status requires an official VIN-level check.

How should duplicate catalog vehicles be handled?

Query each distinct make, model, and model year once, then join results back to matching listings.

Related

Try it yourself

100 free credits, no credit card.

About 30 real searches. Add the MCP to Claude or Cursor in two minutes.