The governed loop

Everything in eavantage comes back to one cycle. Understanding it is enough to understand how the product thinks.

1. Discovery material goes in

You upload what the client has — strategy documents, process maps, application inventories, spreadsheets, interview notes, and diagrams (Visio, BPMN, draw.io, ArchiMate). eavantage reads all of it. See Uploading artifacts for formats and what happens to each.

2. The AI proposes candidates

The AI extracts capabilities, processes, applications, data objects, and the relationships between them. Business capabilities are anchored to the APQC Process Classification Framework, so the structure stays consistent from one engagement to the next. Every candidate carries a link back to the exact place in the source it came from — you can always see why the AI proposed something.

A candidate is a proposal. It is not in the model yet.

3. Candidates are reviewed by risk

Each candidate is risk-tiered by its confidence, how many independent sources support it, and what kind of element it is:

• High-impact elements — gaps, recommendations, objectives — and anything a reviewer flags or that sources disagree on go to individual review.
• Mid-confidence extractions go to a batch review queue.
• Only high-confidence inventory facts that three or more independent structured sources already agree on are committed automatically — and those are logged, flagged to the engagement lead, and can be rolled back.

The AI never decides which findings matter. See Reviewing AI candidates.

4. Approved elements become authoritative

Approved elements enter the model of record — the authoritative picture of the client’s architecture. One software component writes that model. Nothing else can, on any code path (the auto-commit path in step 3 goes through that same writer). Combined with per-client data isolation and a complete audit trail, that single-writer rule is what lets you and your client stand behind the result.

The loop repeats

As more material comes in and the picture fills out, the loop runs again. It is the engine of Phase 1 (Client Discovery) and it keeps turning through target design as recommendations and TO-BE processes are proposed, reviewed, and committed.

---

Why it’s built this way

A client’s reviewers will ask two questions about any architecture an AI helped produce: how do we know this is right, and what did the AI actually touch?

The governed loop is the answer. Because every element was approved by a named person, and because only one component can write the model, you can show exactly which elements are authoritative, who approved each one, and what the AI proposed versus what a human decided. That is the difference between an AI-assisted engagement you can defend and one you can’t.

This is covered for a prospect audience on How it works and Security & data handling.