Sheaf
Knowing what is missing
Most systems report what they found and stay quiet about the rest. The rest is usually what is holding the file up.
The question nobody answers
Ask a document pipeline what it extracted and you will get a thorough answer. Ask it what should have been there and was not, and most of them have no way to represent the question. Extraction is a function of the documents that arrived; completeness is a function of the documents that were supposed to.
In practice this is the difference between a system that saves a reviewer time and a system that changes when a file can move. Nobody is blocked because a value took thirty seconds to find. They are blocked because the fourth document never came and nobody noticed for two days.
Requirements, not fields
Sheaf works against a statement of what a complete file looks like — the requirements, expressed once, at the level of the thing you need rather than the layout it might arrive in. An arriving document is matched to the requirements it could satisfy, and filing it against one is what marks that requirement met.
Because requirements are stated independently of layouts, a requirement can be satisfied by a document nobody anticipated, and an unsatisfied requirement is nameable even though no document exists to point at.
Three kinds of nothing
- Not supplied The requirement has no document filed against it. Someone has to send something.
- Not stated The document arrived and does not answer the question. Chasing a replacement is pointless; the form does not carry it.
- Not applicable The requirement does not apply to this file. Nothing is wrong and nothing is owed.
An empty field collapses all three into one ambiguous signal, and they lead to three completely different next actions. Keeping them apart is most of the value.
Coverage is derived, never stored
Coverage is computed from what is filed right now, not written down when a run finished. A stored percentage is a claim about the past that goes stale the moment a document is re-filed, a boundary is moved, or an approval is withdrawn — and a stale completeness number is worse than none, because people act on it.
One derivation, everywhere. The counter at the top of the room, the list of gaps, and the register all answer the same question with the same arithmetic, so they cannot disagree.
Who else answers this
V7 Go is the one product we know of that goes at completeness directly. Its Missing Document Detection agent compares a repository against a requirements checklist and reports which documents are absent, and its validation agent checks field completion and consistency across documents, with citations. If you are evaluating this capability, evaluate them too.
The design difference is where the requirement list comes from. Theirs is authored and curated by you in a knowledge base, and files are imported to be checked against it. Sheaf derives the requirement set for an application and matches documents to it as they arrive, which is the better fit when the documents come from people who do not know your checklist exists, and the worse fit when your requirements are stable and worth curating centrally.
What the report says
Not a score. The unsatisfied requirement, by name, in the same language your checklist uses — so the output of the system is a sentence you can forward to whoever owes you the document, rather than a dashboard someone has to interpret first.