Sheaf
What the work requires

Audit trail and reviewer decisions

Every automatic outcome points back at the human decisions behind it.

A value walks back to a page

The first half of an audit trail is evidence: any extracted value resolves to a file, a page and a span, and can be opened rather than argued about. That holds for values a person never looked at, which is the point — the trail is not something a reviewer creates by touching a record, it is a property of every value in the system.

A decision records what it rested on

The second half is provenance. When a person overrules an outcome, what is recorded is not merely the new value but the shape of the disagreement.

  • Actor Who decided. A person, a workflow, or an agent — recorded as what decided, which is a different question from who was permitted to.
  • Proposal What the system had suggested before the decision, kept alongside the outcome rather than overwritten by it.
  • Scope What the decision applies to: this value, this document, or this class of document from this counterparty.
  • Departure How far the decision moved from the proposal, which is what makes a pattern of corrections visible rather than anecdotal.

A rule points at the decisions that made it

When the same decision is made a second time and corroborated, it stops being a question and becomes a rule scoped to your organization. That rule is not an opaque preference: it carries the decisions that produced it, so the answer to why did it do that is a list of the times your team did that, with dates and actors.

The distinction

Most systems learn in a direction you cannot inspect and cannot undo. Here, a rule is a record of your own decisions, and it comes back out as easily as it went in.

Reversible, by design

A rule that cannot be withdrawn is a liability, because the desk that created it changes: a policy is revised, a counterparty is dropped, a reviewer's shortcut turns out to be wrong. Withdrawing a rule restores the prior behaviour and leaves the withdrawal in the record, so the trail explains both why the system used to do something and why it stopped.

What an auditor can do without asking you

  • Take any value in a completed file and open the page it was read from.
  • See whether a value was accepted automatically or decided by a person, and which.
  • Follow an automatic outcome to the rule behind it, and the rule to the decisions behind that.
  • Establish when a behaviour started, because rules carry the date they were promoted.

Put real document mess to the test. Send the bundle that currently ruins someone's afternoon, exactly as it arrived.

The rest of the site

Everything Sheaf claims, in writing

Document work is bought on specifics, so the specifics are on their own pages: what we do that a neighbouring category does not, one page at a time.

Against the alternatives

Sheaf vs LandingAI ADE

Excellent composable primitives with real visual grounding. Sheaf is the finished desk those primitives would need to be assembled into.

Sheaf vs Reducto

Agentic Deep Split and grounded citations. Sheaf adds the application, the approval and the coverage arithmetic on top.

Sheaf vs V7 Go

The other product that goes at completeness. Theirs checks a checklist you curate; Sheaf derives the requirements and matches documents as they land.

All comparisons

Ten head-to-head pages and one table — LandingAI, Reducto, V7 Go, Azure, Google, Textract, Instabase, Hyperscience, Rossum and the general models.