HOW IT WORKS
Collected at the authority. Bound to the document. Open to inspection.
Regulated work carries constraints a default configuration doesn't reach. We configure the core platform around your requirements, wired to your own information sources, and adapted to the geographies and workflows you actually operate in.
How we build
Language models now read long documents well enough to do real research work. That is exactly why the standard has to come from somewhere else: from the lawyers, investigators, regulators and auditors whose findings have to survive a hostile reader. Three principles follow.
The engine reads. A person decides.
We design around human-final workflows. The engine compiles, extracts and sorts, so the professional spends their judgment on the files and the findings that need it.
Nothing is asserted that cannot be shown.
Every output is grounded in a retrieved record: a filing, a docket, a registry entry, a guidance, a claim. If the engine cannot point to the document, it does not make the statement.
You can see how a claim got there.
Each finding carries the path behind it, from the sentence back through the document to the authority that published it. A reviewer who was not in the room can follow that path and arrive where we did.
PROVENANCE
What you are trusting, and why you can check it.
A dossier is a set of claims, and each one stands on documents published by an authority. Drawn out, that is a graph rather than a list, and it shows the thing a list hides: which claims rest on more than one independent document, and which rest on exactly one.
Provenance · the path behind one sentence
Every claim is a path you can walk back.
two documents, two filings, one answer
Two paths from two authorities corroborate. One path is still one path, and the file says so rather than rounding it up.
WHAT EVERY DEPLOYMENT INHERITS
The guarantees do not vary by customer.
Whatever the domain and wherever it runs, the same properties hold, because they are how the engine is built rather than options on it.
Every claim, bound to its source
Each value links to the document, page and line it came from. Nothing is asserted that cannot be shown, and a reviewer confirms instead of re-reading.
Defensible months later
The sources behind a finding are held as retrieved, so when someone questions the file a year on you can show what you knew, when you knew it, and what has moved since.
Current at the source cadence
We sync each authority on the schedule it publishes: weekly for clearances, near-daily for recalls and warning letters, and as each county reposts its rolls.
Coverage you can check
A jurisdiction-by-jurisdiction manifest of what we hold, how fresh it is, and where a record is stripped, gated or withheld by statute. A gap is stated, never papered over.
Search inside every filing
Most public databases query a handful of fields. We index every word inside every document, so the predicate, the prior owner or the parallel case is one search away.
Configured, not rebuilt
You get the same platform we run for everyone, configured to your sources, formats and constraints, deployed in your environment or ours. Ask us about your environment and we will show you how it would run.
Some of the foundation is public. Go read it.
FDA's eSTAR template is a PDF form, and PDF forms in government use still run on XFA—a format most libraries gave up on. Handling it correctly is not incidental to submissions work; it is the floor. So we wrote the library, and we published it.
Open source · MIT licensed
pdfer
Pure Go, zero CGO, zero external dependencies →
The claims we make about submissions rest on this. You don't have to take our word for how it works.
Read the source on GitHubHave a constraint the standard configuration doesn't cover?
Tell us your environment and requirements. We will show you how we would deploy.