US Recall Change Audit — FDA / CPSC / NHTSA

Read official U.S. recall records from openFDA (food / drug / device enforcement), CPSC (saferproducts.gov), and NHTSA (vehicle recalls by make / model / year). Normalise them into a flat schema, and optionally report changes this Actor observed since your previous snapshot.

This Actor:

• uses official HTTP GET APIs only (no proxies, no social/e-commerce scraping, no image mirroring)
• does not give safety, medical, or legal advice
• does not promise real-time coverage or zero missed recalls
• does not merge recalls across agencies into one “same recall”
• for NHTSA, only queries the vehicle combinations you provide (not a full national vehicle-recall feed)

What you get

Each dataset item includes:

Field	Meaning
source	fda_food / fda_drug / fda_device / cpsc / nhtsa
sourceId	Stable id inside that source (recall_number, CPSC RecallID, or NHTSA campaign+vehicle key)
officialUrl	Link to the official notice when available
publishedDate / sourceUpdatedDate	Dates from the source (sourceUpdatedDate is null when the source does not provide one)
retrievedAt	When this Actor read the record
recordHash	Hash of the comparable factual fields
changeType	snapshot, new, changed, or baseline (first delta run)
changedFields / oldValue / newValue	Populated for changed
sourceStatus	Per-record source status (ok when built from a successful read)
plus	title, summary, organization, productNames, classification, status, hazard, remedy, vehicle fields, attribution

Street addresses, e-mail fields, and CPSC ConsumerContact phone text are not output. Official remedy / summary text is kept as published by the agency.

Modes

snapshot

Fetch and normalise records for the date window. Every output row has changeType=snapshot and is billable as a recall-record event (when pay-per-event is enabled).

delta

Compare the current fetch to your baseline and output only new / changed rows.

• First run with no baseline: establishes a baseline (changeType=baseline). Those rows are written for transparency but are not charged as recall-record.
• Baselines are customer-isolated. Provide one of:
  • baselineRecords — paste prior records / snapshot JSON in the input
  • baselineDatasetId — a dataset id from your previous run
  • snapshotStoreName — a named key-value store you choose; it is created under the Apify account that runs the Actor (not a shared author store)
• This Actor only reports changes it observed between two of its own snapshots. It does not reconstruct official historical edit times. Narrowing the date window can miss edits to older notices.

Input

Field	Default	Notes
mode	snapshot	snapshot or delta
sources	["fda_food","cpsc"]	Add fda_drug, fda_device, nhtsa as needed
lastNDays	30	UTC window ending today; ignored if publishedFrom is set
publishedFrom / publishedTo	—	Inclusive YYYY-MM-DD
vehicles	[]	Required for nhtsa; max 10 {make,model,modelYear} objects
maxItems	200	Cap across all sources (1–10,000)
baselineRecords	—	Delta baseline in-input
baselineDatasetId	—	Delta baseline from your dataset
snapshotStoreName	—	Optional named store for saving/loading snapshots
snapshotKey	recall-snapshot	Key inside that store

Example (CPSC + FDA food, recent window — matches the Store example intent):

{
  "mode": "snapshot",
  "sources": ["fda_food", "cpsc"],
  "lastNDays": 30,
  "maxItems": 20
}

Example with NHTSA (vehicles required):

{
  "mode": "snapshot",
  "sources": ["cpsc", "nhtsa"],
  "lastNDays": 365,
  "maxItems": 50,
  "vehicles": [
    {"make": "Honda", "model": "Accord", "modelYear": "2018"}
  ]
}

Output

Dataset items as above. The run also writes RUN_SUMMARY to the default key-value store with diagnostics per source (ok, partial, rate_limited, timeout, error, skipped), counts, baseline info, and a disclaimer.

Failures are not reported as “no recalls”. If every requested source errors, the run fails and RUN_SUMMARY.diagnostics explains timeout / rate limit / HTTP errors and lastSuccessAt when available.

Pricing (pay per event)

Event	Price
Actor start (apify-actor-start, platform default, per GB of memory)	$0.00005
recall-record — one billable normalised recall or observed change written to the dataset	$0.005 ($5.00 per 1,000)

Not charged as recall-record: baseline establishment rows, unchanged delta rows (omitted), and error diagnostics. The Actor stops cleanly when the run’s max total charge is reached. Default dataset per-item platform charging is not used as a second fee on top of recall-record.

Default run resources: 256 MB memory, 300 s timeout.

Attribution and terms

• openFDA: general content is CC0; third-party material may be excepted. See openFDA terms and authentication / rate limits (anonymous: 1,000 requests/day/IP, 240/minute).
• CPSC: recall notices may be copied with attribution and without implying endorsement. See CPSC API info and privacy / reuse policy. This Actor attributes CPSC and does not use CPSC badges or logos.
• NHTSA: public vehicle-recall API by make/model/year. Coverage equals the vehicles you list only.

Limits and honesty

• Anonymous openFDA quotas apply; on HTTP 429 the Actor backs off / stops that source and records rate_limited — it does not rotate IPs to bypass limits.
• CPSC / NHTSA exact quotas are not published here; requests are rate-limited conservatively.
• Data may be delayed relative to the agencies’ own sites.
• Fuzzy brand / UPC matching is not treated as a definitive hit across sources.

Local development

python -m venv .venv
.venv/bin/pip install -r requirements.txt
.venv/bin/python -m unittest -v
# optional tiny live smoke (few requests only):
US_RECALL_LIVE=1 .venv/bin/python -m unittest -v tests.test_recalls.LiveSmokeTests