food-recall-radar

Food Recall Radar

Overview

Food recalls happen constantly — thousands per year in the US alone — but almost nobody hears about the one that affects their pantry. News covers outbreaks (Listeria in ice cream, Salmonella in eggs, undeclared peanuts in a cookie), not the 200 quieter recalls of store-brand products. The FDA's own estimate is that recalls reach consumers only a fraction of the time, and contaminated lots sit in kitchens for weeks. The gap isn't data — recalls are public within days — it's matching: nobody has the patience to cross-reference a scrolling recall list against their own groceries.

This skill closes that gap with a personal pantry inventory + live recall matching:

1. Pantry inventory — record what you buy (brand + product, optionally UPC and lot). A one-time 15-minute pantry scan, then 20 seconds per grocery run.
2. Live matching — pulls ongoing recalls from openFDA's food enforcement database (no API key required for light use) and fuzzy-matches against your items: brand containment, product-token overlap, sequence similarity.
3. Lot intelligence — extracts lot/code/date patterns from the recall notice and tells you exactly where to look on your package, instead of making you re-read the whole notice.
4. Ranked actions — Class I (reasonable probability of serious harm) first: check the package / discard / return for refund.

Everything is offline-first: pantry is local JSON, recall data can be cached with --offline, and the script is pure stdlib.

When to Use

• Heard about a recall/outbreak — "wasn't there a spinach recall? did we buy that brand?" → match
• Weekly household audit — 30 seconds, catches the recalls you never heard of → audit
• After a grocery run — add new items in one command → import-receipt
• Pantry reorganization / moving — build the initial inventory → add
• Vulnerable household — pregnant, elderly, under-5, immunocompromised, allergies: undeclared allergen recalls are Class I for you even when rated II for the general public
• Don't use for: pet food (FDA maintains a separate animal-food enforcement file), drugs, or devices — different openFDA endpoints; and never for diagnosing illness — that's a doctor.

Commands

# Record what you own (UPC optional but makes matching near-exact)
python3 scripts/recall_radar.py add --brand "Pillsbury" --product "Golden Layer Biscuits" \
    --upc 018000000000 --lot "K1234" --notes "bought 2026-08-30, fridge"
python3 scripts/recall_radar.py add --brand "Simple Truth" --product "Organic Baby Spinach 5oz"

# Paste a whole receipt / shopping list at once: "Brand Product" per line
python3 scripts/recall_radar.py import-receipt
Pillsbury Golden Layer Biscuits
Simple Truth Organic Baby Spinach 5oz
^D

# Check your pantry against LIVE ongoing recalls (openFDA)
python3 scripts/recall_radar.py match

# Broad weekly audit — active recalls + category heuristics even without brand hits
python3 scripts/recall_radar.py audit

# Use a cached JSON file instead of the network (offline / rate-limited)
python3 scripts/recall_radar.py match --offline --data references/sample-recalls.json

# Manage inventory
python3 scripts/recall_radar.py list
python3 scripts/recall_radar.py remove --brand "Simple Truth" --product "Organic Baby Spinach 5oz"
python3 scripts/recall_radar.py report          # full history of matched recalls

Pantry lives at ~/.pantry.json (--file to override). An openFDA API key is optional (--api-key or OPENFDA_API_KEY env) — without one you get a lower rate limit, which a weekly household audit never approaches.

How Matching Works (so you can trust or override it)

Every ongoing recall is scored against every pantry item on three signals:

Scores above the threshold print with a VERIFY marker — a human looks at the package. UPC equality alone (when both sides have one) is a near-exact hit marked MATCH. The tool deliberately over-notifies: a false "go check the fridge" costs 10 seconds; a missed Class I Listeria exposure costs a lot more. Never discard food purely on the script's say-so — confirm against the linked recall notice, which is printed with every hit.

Reading a Hit

[RISK 1] Class I  ── brand: Pillsbury / product: biscuits
  recall:     Soft biscuits recalled over Listeria monocytogenes
  firm:       Generic Foods Co.
  reason:     Listeria monocytogenes
  dates:      2026-08-01 → ongoing
  lot intel:  look for: lot codes starting 'K1', best-by 09/2026
  action:     DO NOT EAT — check package; return for refund or discard
  notice:     https://www.fda.gov/safety/recalls/...

Class I = reasonable probability of serious adverse health consequences (act now, especially for vulnerable household members). Class II = remote probability / temporary harm. Class III = unlikely to cause harm. Distribution dates tell you whether your purchase window overlaps.

Common Pitfalls

1. Trusting brand names literally. Store brands are manufactured by third parties: "Simple Truth" spinach may be recalled under the co-packer's name. That's why audit also runs category heuristics (product-type words without a brand hit → WATCH).
2. Ignoring UPC check digits. If you type a UPC by hand and get a MATCH on a wrong-digit variant, verify the full 12-digit string against the package before discarding anything.
3. Treating "no results" as "safe forever." Recalls are matched against ongoing enforcement reports. Run audit weekly; newly announced recalls appear within days.
4. Discarding without reading the notice. Many recalls affect specific lots/plants/dates only. The printed notice URL states exactly which packages are affected — read it before throwing food away or demanding refunds.
5. Rate limits. Without an API key openFDA allows modest request volumes; a household runs 1–2 requests per week. If you script this hourly, get the free key and set OPENFDA_API_KEY.
6. Allergen recalls and severity ratings. A Class II "undeclared milk" is Class I for your household if someone is anaphylactic. Severity flags are population-level, not personal.

Verification Checklist

• Pantry has at least the staples you'd be upset to hear were recalled (produce, dairy, frozen, pantry goods)
• match runs clean against live openFDA (or --offline with cached data)
• Every VERIFY/MATCH hit checked against the printed FDA notice URL
• audit scheduled weekly (add it to whatever runs your household chores)
• New grocery trips appended via import-receipt — 20 seconds, not a re-scan

One-Shot Recipes

"There was a recall on the news about ice cream"

python3 scripts/recall_radar.py add --brand "Nice!" --product "Vanilla Ice Cream 48oz"
python3 scripts/recall_radar.py match

Weekly Sunday audit (offline-first)

python3 scripts/recall_radar.py audit --api-key "$OPENFDA_API_KEY"

Testing the matcher without network

python3 scripts/recall_radar.py add --brand "Tasty Blend" --product "Frozen Strawberries 12oz"
python3 scripts/recall_radar.py match --offline --data references/sample-recalls.json

Vulnerable household member (pregnancy / chemo / toddler)

python3 scripts/recall_radar.py audit   # weekly, minimum — Class I Listeria hits matter most