# How do you keep EA factsheets current with AI descriptions and relation recommendations?

> Canonical: https://decisionpen.com/blog/how-to-keep-ea-factsheets-current-with-ai-descriptions-and-relation-recommendations  
> Published: 2026-10-27 · DecisionPen · Living EA series (D7f)  
> Related docs: /docs/incremental-modelling, /docs/architecture-decision-workspace, /docs/enterprise-ask, /docs/beyond-traditional-ea

## Summary

Keep factsheets current by letting AI draft per-entity descriptions from bound evidence and map context, and recommend relationships that no single document states — with humans curating both. The map's connections evolve as evidence extends, confirms, or contradicts existing edges. Builds on "why EA factsheets miss the How and Why" (E3) and "how to enrich factsheets with unstructured documents" (S5b).

## Why factsheets go stale or stay thin

Descriptions, relationships, and rationale are costly to write by hand for hundreds of entities, so the work is deferred: names without meaning, links true only at workshop time, How/Why elsewhere.

## AI-generated entity descriptions

1. Gather evidence and neighbours for the entity.
2. Draft a concise, grounded description in house style.
3. Show provenance.
4. Curate: accept, edit, reject.

Enrichment pulls facts in; description generation makes them readable.

## Relation recommendations

A proposed edge between two entities with the evidence behind it — typed by the metamodel, evidenced, ranked, reversible (rejections are remembered). A candidate until a human accepts it.

## Graph edges no single document knows

Relationships are often implied across sources: a queue in one document, its publisher in another, a platform in an inventory row, a retirement decision elsewhere. The graph can connect them; per-document summaries cannot.

## How connections evolve

New evidence can extend, confirm (raising confidence), contradict (creating a review item, not a silent overwrite), or retire edges. This is incremental modelling in practice and improves agent reasoning over the map.

## Keeping humans in control

Everything is a proposal with evidence; review by priority queue; measure description coverage, accept/reject rates, edges added per week; stewardship stays with the architecture owner.

## How DecisionPen solves this

Architecture Decision Workspace: AI descriptions per entity with provenance; typed, evidence-backed relation recommendations; human curation; living map feeding Enterprise Ask and generated artefacts. Trial: https://decisionpen.com/trial

## When is DecisionPen the wrong tool?

Unattended auto-published factsheets with no review; runtime-only asset inventory; drawing canvas.

## FAQ

**How can AI keep EA factsheets up to date?**  
By drafting per-entity descriptions from bound evidence and recommending relationships, with human curation before anything becomes authority.

**What is a relation recommendation?**  
A proposed, typed edge between two entities with supporting evidence, treated as a candidate until accepted.

**Why do some relationships appear in no single document?**  
They emerge when several documents are read together; a curated graph can hold and recommend those connections, while per-document summaries cannot.

**How do connections evolve over time?**  
New evidence can extend, confirm, contradict, or retire edges; contradictions become review items rather than silent overwrites.

**Is this fully automatic?**  
No. AI proposes; an accountable steward curates.

**When is DecisionPen the wrong tool?**  
When you want unattended publication, only runtime inventory, or a drawing canvas.

## Related posts

- https://decisionpen.com/blog/why-ea-factsheets-miss-how-and-why  
- https://decisionpen.com/blog/how-to-enrich-ea-factsheets-with-unstructured-documents  
- https://decisionpen.com/blog/how-do-ai-agents-interrogate-and-deep-reason-over-an-enterprise-architecture-map  
- https://decisionpen.com/blog/how-to-generate-adrs-blueprints-and-diagrams-from-living-architecture-context