Your regulation has a decimal. FDA's watch doesn't.

Pull the classification record for product code IRT— powered heating pad — and FDA gives you a precise regulation: 21 CFR 890.5740, Class II, exempt from premarket notification, sitting in the Physical Medicine Therapeutic Devices subpart of Part 890. That citation is exact down to the decimal. It names one device type, not a neighbourhood of them. Set up a watch on that same regulation, expecting it to catch the Federal Register notice the day FDA amends it, and the watch has a gap built into it before a single document arrives.

Two different units of measurement

A device classification regulation is written at section granularity because that is what a classification is: one section, one device type, one set of controls. Part 890 alone runs to dozens of sections — catheters, exercisers, traction equipment, heating pads — each with its own three- or four-digit suffix and its own separate finding. The Federal Register, covering the part where a proposed or final rule actually changes something, does not carry that same resolution. Its structured cfr_references field — the one a monitoring tool actually reads — records a title and a part: “21 CFR 890,” never “21 CFR 890.5740.” The API’s own CFR search parameters stop at the same place: title and part, no section. A rule that reclassifies heating pads specifically and a rule that touches an unrelated corner of Part 890 file under the identical citation.

What that does to a watch built on the wrong key

Configure a monitor to match on the full citation — “21 CFR 890.5740,” verbatim — against a source that only ever publishes “21 CFR 890,” and it matches nothing, ever. Not because the rule wasn’t relevant. Because the two records were never going to collide on a string comparison to begin with; the granularity mismatch is permanent, not occasional. The practice this argues for is matching at the part — the level the source actually publishes at — and then reading, not filtering, every part-level hit to see whether it touches your specific section. That second step is unavoidable, and it is also the entire value of the exercise: a part can move for reasons that have nothing to do with your device.

The trade this makes, stated plainly

Matching at the part is deliberately biased toward showing you more than you need, because the alternative — matching at the section and trusting a Federal Register notice to cite it that precisely — silently shows you nothing, including the one week it mattered. FDA Radar’s own product-profile join runs on exactly this rule: a product’s profile carries its full classification regulation, decimal included, but the “why this affects you” line a match renders is deliberately coarser — “Cites 21 CFR 876 — your regulation” is the literal join key shown, part number only, never the section — because the part is the level the underlying document actually supports, and showing a more precise citation than the source justifies would be a claim the join can’t back up.

Whether a given part-level hit is actually about your device is a judgment call for whoever reads it, against the rule’s own text — this is a description of where a citation-granularity mismatch lives, not a determination about any specific regulation. But it is worth checking, once, which granularity a watch you rely on actually matches at. A regulation number with a decimal in it is not the same key as the one the source publishes, and a filter built on the wrong one fails exactly the way silence does: without telling you.