FDA Import Alerts · implementation

Screening supplier names against FDA Import Alerts: the name traps

An Import Alert screen is a name comparison, and FDA's lists spell names the way whoever typed them spelled them. Five of those spelling habits will make an automated screen answer “not listed” about a firm that is listed. They are all visible in the published data, and none of them is exotic.

1. The legal form is written a different way than you write it

A firm you know as Claridges Organic Ltd is published as Claridges Organic Limited. Compared as strings, those are different names. Screened on 2026-09-17 through a matcher that compares each name with and without its trailing legal form, the pair is an exact match at score 1 on the red list.

This is not a rare edge. Over the generation of FDA's lists this page was written against, 1,740 firms on FDA's published lists carry a legal form in their listed name (measured 2026-09-17) — and 322 of those were missed by a screen that left the form off, before the fix. The forms are not only the English ones:

• Ostrea → Ostrea d.o.o. (Croatian d.o.o.), exact at 1
• SUNEX → SUNEX S. de R.L. (Spanish S. de R.L.), exact at 1

What to do: compare each name twice, once as written and once with the trailing legal form removed, and keep the better score. Do not strip the form permanently — it is part of the identity, and two different firms can share everything but their form.

2. The spaces are somewhere else

Transliterated names arrive spaced by whoever typed them: GUANG ZHOU HUA FENG on one alert, Guangzhou Huafeng from your supplier's invoice. A substring pre-filter never even offers that pair to the scorer, because neither token appears in the other string. The fix is to compare a spaceless form of both sides as well, and to admit candidates on that basis before scoring.

Screened on 2026-09-17, Shandong Ocean Fisheries returns no_match — with 67 entries disclosed as read and scored below the match floor. That number is the point of this trap: the screen looked at 67 names close enough to score, and a screen that reported only “no match” would have hidden all of them.

3. A letter is not the letter you think it is

Turkish ı (dotless i), Danish ø, Polish ł, German ß: naive normalization deletes them, turning a name into a different name. Unicode's NFKD decomposition does not decompose these at all, so a pipeline that strips combining marks will erase them silently and leave a gap where a letter was. Fold them explicitly, in both directions, and compare the folded forms.

4. A hit on the Green List is the opposite of a problem

Each alert publishes a Red List (firms meeting the criteria for detention without physical examination), a Yellow List (in FDA's words, firms “subject to intensified surveillance”) and a Green List (firms that have met the criteria for exemption from that alert's detention). A screen that reports “found on Import Alert 16-131” without saying which list can turn an exemption into an alarm. Each response above carries list_type for exactly that reason — and a Green List entry exempts a firm from that specific alert only, never from any other.

5. The alert that names no firm at all

Some Import Alerts detain every shipment of a product from an entire country, some only in a stated season, unless the firm appears on that alert's Green List. Your supplier can be covered by an alert while appearing on no list in it. No firm-name screen can answer that question, including this one: it is a country-and-product question, and the honest thing for a screen to do is say so on every response rather than let a clean name check imply clearance. Check the alerts for the country and product directly at FDA's Import Alert list.

What a screen should tell you when it finds nothing

Four of the five traps above end the same way: an answer that looks clean. That makes the response contract as important as the matcher — how many entries were read and scored below the floor, whether anything was withheld, whether any part of the screen was incomplete, and how current the data is. Those fields, and the seven questions to ask of any screening response, are in What a screening API does not show you.

One more trap, and this one we have not solved

Every trap above is one a matcher can be built to handle. A firm name typed without its spaces is not. A compacted string is one token, so it shares nothing with the tokens of the listed name, and what is left is an edit-distance comparison a short string cannot win. We measured it on 97 real listed entities: a short name finds its record 98% of the time spelled with spaces and 4% without, and half the misses score below the floor where a near miss would be reported at all. Every widening of the matcher we measured bought back a fraction of those and created new false matches between genuinely different companies, so the honest answer is to screen the spaced legal name. The measurement, the data and the three rejected fixes: a company name without spaces misses the screening list.

The responses on this page are unedited live captures made on 2026-09-17 against FDA Import Alert Screening, which does all five of the things described here; every figure is parsed out of that capture when the page is built. FDA's lists are public data published at accessdata.fda.gov, and the firms named above are organizations on those published lists.