Assess the impact of retiring a system

Goal

Before a system is retired or replaced, find everything in the model that depends on it, who owns what is affected, how critical it is, and which diagrams show it. For an architect preparing a retirement or replacement decision. You end with an impact report in business terms that states how far the search went, and, only if you ask, the system marked for retirement in the model.

Where it runs

This scenario needs the MCP server. Computer use covers its UI steps.

Before you start

• The start-of-session check done. The scenario reads only, unless you ask for the system to be marked in the model at the end.
• The system and the things that depend on it are elements, and the dependencies are connectors. A dependency recorded only in a note, a tagged value or a diagram's layout is not found.
• You know which connector types count as a dependency in your modeling language. In Westbrook, Uses and Flows connectors, and realizes connectors to requirements and services.
• Optional: the tagged values that say who owns an element and how critical it is. In Westbrook, businessOwner, technicalOwner and criticality.

This scenario puts Trace dependencies to work on one decision. Where that scenario answers a single dependency question, this one assembles the answer a retirement decision needs: what depends on the system, what it depends on, who is affected, and where it is drawn.

Steps

1. AI with computer use; otherwise you If EA is not running or the model is not open, start EA and open it. AI Power Tools ea-start-here runs the start-of-session check.
2. You Name the system, and confirm the element when several share the name. The AI shows each candidate with its package path.
3. You Agree the traversal rule before it runs: which connector types count as a dependency, and how many hops to follow. The AI proposes one from the connector patterns the model actually uses (summarize_connector_patterns).
4. AI Power Tools ea-model-hygiene traces upstream, against the arrows, to find everything that depends on the system (trace_connectors). The default direction is downstream, so the AI asks for upstream explicitly.
5. AI Power Tools ea-model-hygiene traces downstream to find what the system depends on: the data, services and vendor systems a replacement must also cover.
6. AI Power Tools The AI reads the guards on both results: whether either was truncated, and how deep each reached. A truncated or depth-limited result is reported as partial, never as complete.
7. AI Power Tools ea-model-hygiene checks where the system is drawn: a direct query for its placements on diagrams, and the composite audit over its package (find_composite_diagram_mismatches), which shows whether it anchors a navigation diagram.
8. AI ea-stakeholder-reporting reads the owner and criticality tags of each affected element (get_element_tags, or one direct query for many elements) and groups the affected elements by criticality and owner. Elements without the tag are counted as unknown, not left out.
9. AI ea-stakeholder-reporting writes the impact report in display terms (get_element_business_view, resolve_display_term): what depends on the system, grouped by criticality and owner; what the system depends on; the diagrams that show it; the traversal rule; and how far the search went.
10. You Decide what to do with the report. If you ask the AI to mark the system in the model, for example by setting its lifecycle tag to the value your standard uses for retirement, it takes a baseline of the package first and asks before the write (create_baseline, set_tagged_value). Deleting the system is a separate step, covered by the pre-delete safety sequence in ea-model-hygiene.

Prompts to use

Opening

Follow-ups

The AI asks before the baseline and before the write.

Recommended models

Ratings are guidance based on the kind of reasoning the skill needs, not measured benchmarks.

Checking the result

• Accurate. For the system itself, a one-hop upstream trace returns as many connectors as the element has incoming dependency connectors, which you can confirm by listing its connectors (get_connectors_for_element_filtered). Spot-check two affected applications in EA's Relationships view.
• Complete. Neither trace was truncated, and the depth each reached is below the depth asked for, or the report says the limit was reached. The connector types followed are named in the report, so a reader can see what was not followed. Elements with no owner or criticality tag are counted, not dropped.
• Nothing unintended changed. The traversals and tag reads change nothing. If you asked for the system to be marked, a comparison against the baseline taken in step 10 shows that one tag changed and nothing else (compare_baseline).

Variations

• Replacing rather than retiring. Name the replacement too, and ask which of the system's dependants are already connected to it (find_paths between each dependant and the replacement). What is left is the migration list.
• A business capability instead of a system. Trace downstream from the capability along realizes and Uses connectors to see which systems support it, then assess each.
• Seen from Prolaborate. Where stakeholders work in Prolaborate, Trace the impact of a change in Prolaborate traces the same question through Analyzer views, one hop at a time.

Limitations

• Only connectors are followed. A dependency that exists only in a note, a tagged value or a package's nesting is invisible, so the report is as complete as the model's connectors.
• The traversal rule is a judgment. Which connector types count as a dependency is your decision, and the report is only as complete as that rule. Following every type in both directions quickly reaches most of a large model and hits the size limit.
• Large neighborhoods hit the node and time limits. A truncated result is a prefix, not an answer. Narrow the connector types or the depth.
• Diagram placements come from a direct query. The AI reports the query it ran.
• Ownership and criticality are only as good as the tags. Missing or inconsistent values are reported as such. Clean up model decay finds tags set on some elements of a type and skipped on others.