The problem
AI assistance becomes hard to govern when consequential relationship changes happen without visible evidence, ownership, or decision history.
What EKOS does
Reviewers claim work, inspect supporting excerpts and confidence, understand the result of approval, and approve or reject supported proposals with notes and visible state transitions.
- Claim and filter supported review items
- Side-by-side excerpts and confidence
- Reviewer notes and visible state transitions
- Stale state when supporting sources change
- Mandatory review for conflict and supersession proposals
How it works
Proposed relationships and near-duplicate merges enter the queue with excerpts from both documents, a confidence score, and the extraction method.
A reviewer claims the item, inspects what approval will change, and approves or rejects it with notes.
The decision updates the underlying graph or canonical-document record, and later automation is not allowed to silently overwrite it.
Regulatory change readings, compliance findings, and user-flagged edges share the same queue with assignment, status, and type-specific context.
Items turn stale when a supporting source changes, and conflict or supersession proposals always require a human decision.
Business outcome
People focus on specific evidence-backed judgments while approved relationships improve the graph, impact reports, coverage views, and evidence packages that depend on them.
Proof and human control
Relationship proposals retain excerpts, confidence, extraction context, reviewer notes, state, and the decision that changed the underlying graph.
Example in practice
EKOS proposes that a new procedure conflicts with an approved policy. A reviewer compares both excerpts, adds a note, and approves the relationship as a reviewed graph fact.
Scope today
Tenant policy can accept sufficiently confident lower-risk proposals automatically. Conflict and supersession proposals always require human review.