What a device classification tool has to check, and where most stop

Classification is a risk judgment about the device type

21 CFR 860.3 sorts devices into Class I, subject to general controls, Class II, where general controls alone aren't enough and special controls close the gap, and Class III, reserved for devices where neither is sufficient to provide reasonable assurance of safety and effectiveness — typically because the device is life-supporting, life-sustaining, or presents a potential unreasonable risk of illness or injury. That determination is made for a generic type of device, not a specific product, and it's set out in the classification regulation for that device type under 21 CFR Parts 862 through 892. A tool that gets a user to the right class has done real, source-mapped work. It hasn't yet told them what to file.

Most Class I and Class II devices don't need a 510(k) — up to a limit

FDA has exempted the large majority of Class I device types, and many Class II types, from the requirement to submit a premarket notification at all. But every one of those classification regulations carries its own limitations-of-exemption section — numbered .9 within that part, for example 21 CFR 862.9 for clinical chemistry devices — and the exemption only holds to the extent the device has the existing or reasonably foreseeable characteristics of devices already on the market in that generic type. A device that has a different intended use, uses a different fundamental scientific technology, or is a reprocessed single-use version of an otherwise-exempt type exceeds the exemption regardless of what class it's in, and a 510(k) is required after all.

Why the exemption limit is the branch, not a footnote

This is the exact place a classification answer collapses two different questions into one confident-looking output. “What class is this device” and “does this specific device still qualify for that class's exemption” are answered by different sections of the CFR, checked against different facts about the device, and a tool that only asks the first question will tell a sponsor with a reprocessed, exemption-exceeding device that they're clear to market, when the regulation says the opposite. The same rule that makes a checklist item trustworthy applies to a tool's output: one claim, one citation, checkable against the source — not a single answer standing in for two separate determinations.

For the devices that do need a 510(k), the classification question ends and a different one starts

A device that needs a 510(k) — because it's Class III, or because it exceeded its exemption — moves to a wholly separate test: substantial equivalence, defined at Section 513(i) of the FD&C Act and applied under 21 CFR 807.100(b), which asks whether the device has the same intended use as a predicate and either the same technological characteristics or characteristics that don't raise different questions of safety and effectiveness. That's a different regulatory question from classification, decided against a different predicate-specific record, and it's the one an Acceptance Review checks for completeness on without ever judging whether the answer is right. A classification tool that hands a user off cleanly at this boundary — you need a 510(k), and here is the specific test it has to pass — is doing its job. One that just says Class II never gets the user to that boundary at all.

A classification tool built to expose this branch explicitly — class, then exemption limits, then substantial equivalence only where it actually applies — runs entirely client-side, in the tools section now in preview; nothing about a device description leaves the browser to produce that answer. If your team has hit an exemption edge case this logic misses, the shelf takes that correction directly.