RecallDB is a normalized dataset of official U.S. federal recall records from CPSC, FDA/openFDA plus FDA Safety Alerts, USDA FSIS, NHTSA and USCG.

Does RecallDB include provenance?

Yes. Every recall carries source_agency, external_id, source_url and source_id. data_sources.csv exposes source endpoint, retrieval time, raw payload size and SHA-256 fingerprint.

Is the public GitHub repo the full dataset?

No. The GitHub repository and Kaggle sample metadata contain only the public sample, docs and starter assets. The full CSV and SQLite dataset is paid and delivered privately after a snapshot purchase or enterprise request.

Official U.S. recall intelligence - v1 public sample

Five federal recall feeds, one offline, provenance-ledgered dataset.

Name: RecallDB U.S. Product Recall Database — Snapshot
Brand: RecallDB
Price: 49.00 USD
Availability: InStock

CPSC, NHTSA, FDA, FSIS and USCG publish recalls for free, in five shapes. RecallDB joins them into one CSV/SQLite model — 129,279 official recalls, 295,319 linked products — with a full provenance ledger on every row. The companion to MechanicDB for vehicle fault-and-recall screening.

Download sample CSV Get the full dataset View sample repo

$49 one-time snapshot · free 200-row sample — for e-commerce sellers, compliance teams and AI agents that need offline recall data

129,279: official recalls
295,319: linked products
5: federal sources
100%: source-linked

Coverage

Built from official federal recall publications only.

No commercial aggregators, no invented rows, no fuzzy provenance. EPA is intentionally deferred until a stable official recall-specific feed exists.

Traceability

Every recall joins back to its source pull.

The paid SQLite release keeps the full archived raw payload. The paid CSV delivery exposes compact lineage through source_id and data_sources.csv, including endpoint URL, retrieval timestamp, raw payload byte length and SHA-256 hash.

recalls.csvsource_id, source_url, source_endpoint_url, retrieved_at

data_sources.csvendpoint_url, raw_payload_bytes, raw_payload_sha256

Paid SQLitefull data_sources.raw_payload for replay and audit

Public sample

Small enough for GitHub, shaped like the full release.

The sample contains balanced rows from all five source agencies and the same export table shapes used in the paid full dataset snapshot, plus NHTSA investigation, service-bulletin and complaint-count rows for the sample's vehicles.

recalls.csv200 sample recalls with source_id lineage recalled_products.csvProducts, vehicles, lots and categories data_sources.csvEndpoint URLs and payload fingerprints hazards.csvControlled hazard taxonomy counts nhtsa_investigations.csvNHTSA defect investigations and the recalls they led to nhtsa_mfr_communications.csvService bulletins and over-the-air updates nhtsa_complaint_counts.csvComplaint counts per vehicle and component

Latest sample rows
Agency	Date	Title	Hazard

Hazards

One deterministic taxonomy across agencies.

CPSC free text, FDA enforcement reasons, NHTSA components and FSIS reasons map into one auditable vocabulary. USCG boating recalls are included, but the Coast Guard stopped showing defect text on its recall pages in August–September 2026, so USCG rows carry no hazard class.

Beyond recalls

NHTSA investigations, service bulletins and complaint counts.

From edition 2026.10 the full snapshot also carries three tables from NHTSA's Office of Defects Investigation, built from the same official bulk files as the vehicle recalls. They join to recalled vehicles by make, model and model year, and every investigation links to the recalls it led to by campaign number.

Complaints ship as counts only: complaint text and complainant details are never published. VINs that identify one vehicle are masked; production breakpoints stay as NHTSA prints them.

InvestigationsPreliminary evaluations, engineering analyses, recall queries and defect petitions, with the vehicles they name and the recalls they led to. nhtsa_investigations

Service bulletinsTechnical service bulletins, service campaigns, warranty extensions and over-the-air updates filed with NHTSA, with their vehicles and components. nhtsa_mfr_communications

Complaint countsConsumer complaints per make, model, model year and component, with crash, fire, injury and death counts. nhtsa_complaint_counts

Access

Two paid ways to get RecallDB data.

The public repository contains only the sample files and documentation. The full CSV + SQLite export is not hosted on GitHub, GitHub Pages, Kaggle, or any free public download.

Dataset snapshot

Full Dataset Snapshot

One private delivery of the current RecallDB export for analysis, enrichment, compliance research, and internal product screening.

$49 one-time snapshot

129,279 normalized recall records
295,319 linked recalled products
CSV + SQLite delivery
Provenance ledger and source links
NHTSA investigations, service bulletins and complaint counts (from edition 2026.10)
Private delivery after request

Buy $49 snapshot

Secure Stripe checkout. The full snapshot is never committed to this repo and delivery is gated by payment verification.

Bundle

Fault & Recall Bundle

RecallDB Vehicle — the NHTSA vehicle & equipment recall cut — together with MechanicDB OEM Complete, its fault-code companion.

$179 one-time · vs $198 separately

RecallDB Vehicle: CSV + Parquet + SQLite
RecallDB Vehicle adds NHTSA investigations and complaint counts from edition 2026.10
MechanicDB OEM Complete: 15,886 codes, 56,561 ranked repairs, 75,055 parts
Bulletin index (bundle only): NHTSA service bulletins that cite each trouble code, with the vehicles they cover
Instant downloads after one checkout
Commercial license on both

Buy the $179 bundle

Instant delivery: both products are under the delivery worker's file-size cap, unlike the 555 MB full snapshot.

Custom work (custom joins, refresh cadence, retail matching, schema work): from $99 — contact the desk

Direct desk: recalldb@dataengineered.io

Request access

Tell us which full-dataset tier you need.

Requests go to the RecallDB desk. We reply with payment and delivery details; the full export is delivered privately, not through a public repository link.