Ichlibitiche/recalldb-product-recalls-sample

RecallDB — U.S. Product Recall Database (Sample)

Full dataset: recalldb.dataengineered.io · $49 one-time snapshot → Buy on Stripe · the same sample on Kaggle

129,279 official recalls · 295,319 recalled products · CPSC · FDA · FSIS · NHTSA · USCG · 100% source-linked

RecallDB is a normalized, provenance-tracked dataset of official U.S. federal product recalls. It joins five official source families into one relational model: CPSC consumer products, NHTSA vehicles, FDA/openFDA plus FDA Safety Alerts enrichment, USDA FSIS food safety recalls, and USCG recreational boating recalls.

Full dataset & documentation: https://recalldb.dataengineered.io/

Public sample vs paid full dataset

This Hugging Face dataset is the free public sample. The full CSV + SQLite export is not published here.

Table	Full dataset	Public sample
Recalls	129,279	200
Recalled products	295,319	428
Source pulls	2,226	14
Firms	21,112	147
Hazard taxonomy	11	11

Files / configs

Each table is exposed as a separate config for the Datasets viewer:

• recalls — balanced sample of recalls from all five agencies
• recalled_products — linked products, vehicles, lots and categories
• data_sources — endpoint URLs, retrieval timestamps and raw payload hashes
• firms — canonical firm names observed in the sample
• hazards — hazard taxonomy with recall counts
• nhtsa_investigations — NHTSA defect investigations naming the sample's vehicles, plus a few with recall links
• nhtsa_investigation_vehicles — vehicles and equipment those investigations name
• nhtsa_investigation_recalls — recall campaigns those investigations led to (exact, padded_000 and unmatched examples)
• nhtsa_mfr_communications — recent manufacturer communications (TSBs, OTA updates, service campaigns) for the sample's vehicles
• nhtsa_mfr_communication_vehicles — make / model / model year each of those communications applies to
• nhtsa_mfr_communication_components — NHTSA components those communications name
• nhtsa_complaint_counts — consumer complaint counts per component for some of the sample's vehicles (counts only, no complaint text)

Quick start

from datasets import load_dataset

recalls = load_dataset("Ichlibitiche/recalldb-product-recalls-sample", "recalls", split="train")
print(recalls.to_pandas().groupby("source_agency").size())

Or with pandas directly:

import pandas as pd

recalls = pd.read_csv("hf://datasets/Ichlibitiche/recalldb-product-recalls-sample/recalls.csv")

Provenance

Every recall carries source_agency, external_id, source_url, and source_id. source_id joins into data_sources, which exposes endpoint_url, retrieved_at, raw_payload_bytes, and raw_payload_sha256. The paid SQLite release also stores the archived raw payload for replay and audit.

Source coverage

Agency	Full rows	Domain
CPSC	8,335	Consumer product recalls
FDA	87,649	Food, drugs, devices, cosmetics, biologics, animal/veterinary alerts
FSIS	1,236	Meat, poultry and egg product recalls
NHTSA	30,362	Vehicle and equipment recalls
USCG	1,697	Recreational boating recalls

EPA is intentionally excluded from v1 (no stable official recall-specific structured feed was identified).

Full dataset access

The full CSV + SQLite export with the complete provenance ledger is available as a one-time $49 snapshot or an enterprise custom request via https://recalldb.dataengineered.io/#pricing

License

Public sample: CC0-1.0 (source records are U.S. federal government publications). Attribution is appreciated for auditability — see https://recalldb.dataengineered.io/SOURCES.md