How to use this safely

What the analysis decides, what it never decides, and where the judgment stays yours.

This tool produces the analysis layer of a 510(k) in seconds — pathway, product code, predicate, equivalence, hazards, timelines, checklists — from the FDA public record. It is built to be relied on correctly, which means knowing where its authority ends.

What it does

• Gives a firm answer wherever an FDA rule or past clearances settle it — and cites which.
• Shows how often it is wrong, answer by answer, with the count it was measured on.
• Says so when nobody has stated a fact (“the predicate’s record is silent”) instead of inventing a comparison.
• Shows the dossier’s provenance and audit ID, and identifies submitted corrections and completed review actions as user-entered records.

What it does not do

• It does not write your test reports, protocols, or labeling — sponsor content is declared, never generated.
• It does not create a Part 11 electronic signature. A typed reviewer name creates a review record; a qualified person still signs through the sponsor’s controlled process before filing.
• It does not guess when your document leaves a fact open — it asks you the one question instead.
• It only knows what FDA has published. If FDA’s data is behind, the dossier says so, and by how many days.

When to distrust it

• Review required means exactly that: we are not sure enough to decide it alone. Take the hand-off brief to a person. We count every case we send to a person as a miss on our side, never as caution.
• Medium — confirm it means the deciding fact is not in your text. Every answer that depends on the product code carries that uncertainty until you answer the one question shown.
• If the fields we read from your document are wrong, everything after them is wrong — that is why they are shown first, for you to correct. A check mark means found in your own text, nothing more.
• Review times are measured ranges, not promises. Trust the range, not the midpoint.

Where your document goes

Two companies outside this one see your document, and they are not the same company with different controls. Three steps:

1. 1OpenRouter receives the full text to pull out the structured fields — and, if the document is not in English, to translate it. Zero data retention is set on every request.
2. 2OpenAI receives the joined device fields, to find similar predicates. That route has no per-request retention setting, so it rests on the account terms.
3. 3This product keeps only the extracted fields on the audit record. This browser may hold an unfinished draft so a failed run is recoverable; a successful analysis deletes that draft and does not save the source document in the caseload. Predicate ranking runs on our own machine; the FDA, eCFR and Federal Register queries carry no text of yours.

Nothing is filed with FDA by this product.

Measurement

To know whether the product does its job, the page records what happens in it: a verdict appeared, a question was answered, a disclosure was opened, a correction was recorded, a sign-off happened. Timings and shapes only — no document text, no field values, no names. The events stay on our server. Nothing is sent to a third party.

We hold no data-processing agreement with either company on your behalf, and no retention commitment you could hold anyone to. This page describes what the software does, which is not a contract. For an unfiled 510(k), whose existence is confidential under 21 CFR 807.95, that is worth saying plainly.