Three letters is too short to search for

Every Federal Register document FDA publishes gets scored against a term list before it reaches a device-monitoring feed at all — a plain keyword match deciding, cheaply and before any human looks at it, whether a notice is worth a human’s time. Cheap is the right instinct. Naïve is not, and the difference showed up in FDA Radar’s own probe output as a drug guidance about antibodies, filed under devices.

The three letters inside the wrong word

On December 3, 2025 FDA announced a draft guidance titled “Monoclonal Antibodies: Streamlined Nonclinical Safety Studies” — recommendations on when a monospecific antibody needs a full toxicology study and when it doesn’t, aimed at drug sponsors, with nothing about medical devices in it. FDA Radar’s Federal Register filter matched it anyway. The reason was three characters long: udi, the acronym for Unique Device Identification and one of the highest-value short terms on the list, is also the middle of the word studies. A substring match doesn’t know the difference between an acronym and a syllable; it just found u-d-i and moved on.

Why the short ones are the landmines

Long terms rarely have this problem — “medical device” or “premarket notification” is not going to turn up by accident inside an unrelated word. Three- and four-letter acronyms are a different case entirely. udi, ivd, and samd are all real, load-bearing terms in device regulation — Unique Device Identification is a defined system under 21 CFR Part 830, not a term worth dropping from a watchlist to avoid this problem — and all three are also ordinary letter sequences that turn up inside English words with no relation to devices at all. Drop them and a monitoring feed misses real UDI and IVD guidance. Keep them and match on substrings, and a monitoring feed starts filing chemistry into the wrong folder.

The fix is boring on purpose

The fix that actually holds is not a smarter filter — it’s a stricter one: match the term only as a whole word, tolerant of a trailing s for the plural, and reject it anywhere it appears as a fragment of a longer word. “UDI requirements for labelers” matches. “Streamlined Nonclinical Safety Studies” doesn’t, because studies is one word, not two. The rule sounds obvious stated this way and is easy to violate by accident — ordinary word-boundary regex syntax is written against word characters, and a term like 510(k), which ends in a parenthesis, breaks it in the other direction. Getting this right took a second pass and a regression test pinned to the exact document that exposed the bug, not a cleverer library.

Miss loud, not quiet

A filter like this is never going to be perfect, and the honest design question is which direction to be wrong in. FDA Radar’s term list is deliberately biased toward catching too much rather than too little: a false positive costs a reviewer a few seconds discarding an irrelevant notice; a false negative means a real device guidance never reached anyone at all. That trade only makes sense because every match still passes in front of a person before it reaches a user — a keyword filter that fed straight through to a subscriber, with nothing reviewing the output, would need to hold a much higher bar for precision than this one does.

Which terms belong on a watchlist, and how wide to cast the net, is a judgment call for whoever is building it, against their own portfolio — this is a description of a failure mode in the mechanism, not a recommendation of which acronyms to track. But the mechanism generalizes past the Federal Register: a saved search, a Google Alert, an RSS filter built on the same three- and four-letter terms this industry runs on will fail exactly the same way, in exactly the same direction, until it is checked against a word boundary instead of a substring.