Dependencies / Impact Analysis (Pro)

manage_dependencies prevents you from breaking things when you change a semantic model. In practice, it answers:

"If I rename this column, what measures will break?" - it finds every DAX expression that references Customers[Region] and tells you exactly what needs updating.
"Can I safely delete this measure?" - it checks whether other measures, calc items, or report elements depend on it and shows you the full blast radius.
"What does this measure actually reference?" - it traces the dependency chain so you understand how a complex measure is wired into the model.

It builds an impact graph across DAX expressions, M expressions, partitions, relationships, sort-by columns, hierarchies, perspectives, calendars, and security metadata.

You don’t need to know tool parameters-ask the LLM for an “impact analysis” and a PR-ready summary, and it will use manage_dependencies behind the scenes.

What to ask the LLM (quick prompts)

“Find everything that depends on measure [Total Sales] (direct + transitive).” “If we rename Customers[Region], what will it impact? Give me a PR-ready summary.” “Show the dependency graph for calc group ‘Time Intelligence’ and highlight risk areas.” “Render a dependency diagram (Mermaid) and also provide a short human-readable summary.”

Concepts (plain English)

Direct vs transitive dependencies

Direct: items that reference the target directly.
Transitive: items that reference something that references the target (the “blast radius”).

Prompt:

“Show direct dependents, then transitive dependents (depth 2).”

Expression vs structural dependencies

Expression dependencies come from code (DAX/M text).
Structural dependencies come from model wiring (relationships, sort-by, hierarchy membership, perspective membership, calendar mappings, etc.).

Prompt:

“Include structural dependencies (relationships, sort-by, hierarchies, perspectives).”

Why results can be “noisy”

Dependency analysis includes fast text matching and heuristics. You can get false positives-especially for generic terms like Date, Value, Amount.

Prompts:

“Use high-confidence matches only.” “Restrict to measures + calculation items only.” “Keep depth to 1 unless needed.”

Recommended workflow (impact-first changes)

Rename safety check (recommended)

“Show direct dependents (depth 1).”
“Expand to transitive dependents (depth 2).”
“Summarize what will likely break and what can be updated automatically vs manual.”
“If approved, apply the rename and re-run dependency check.”

Copy/paste prompt:

“Before renaming Customers[Region], show direct + transitive dependents (depth 2). Summarize impact and propose a safe change plan.”

Delete safety check (even stricter)

“Before deleting [Legacy Metric], show all dependents (depth 3) and tell me whether deletion is safe. If not safe, propose deprecation steps.”

Cleanup/refactor work (design improvements)

“Find measures depending on deprecated columns and propose a refactor plan grouped by table/domain.”

Output formats you can ask for

Depending on your workflow, ask the assistant to render results as:

Plain English summary (stakeholders)
Markdown tree (PR descriptions)
Mermaid diagram (visual impact graph)
CSV edges/nodes (import into graph tooling)

Copy/paste prompts:

“Generate a PR-ready ‘Impact analysis’ section as a Markdown tree for renaming [Total Sales].” “Render the dependency graph as a Mermaid flowchart and also provide a plain-English summary.” “Export dependency edges as CSV so I can analyze them externally.”

Locked-down environments (mode behavior)

Mode	Availability
Full mode	AvailableAvailable
Read-only mode	AvailableAvailable
Browse-only mode	Not availableNot available

Notes: Impact analysis for renames/deletes/refactors; depth can get large.

Learn more about modes and restrictions.

Fallback in restricted environments:

“Without dependency tooling, use list_model expression search to approximate impact and give me a manual verification checklist.”

Practical tips (consultants love these)

Always run this before rename/delete.
Start with depth 1 and only increase when needed (depth 3–4 can get large).
Narrow scope to keep results actionable (“measures only”, “DAX only”, “exclude metadata edges”).
Re-run after major edits (especially calc groups and relationship changes).

Troubleshooting

Ask for a fallback: list_model expression search + a manual verification checklist.

Ask: “Tell me what setting controls dependency index build timeout and what the current value is.”

Ask: “Restrict types (measures only), use high-confidence matches only, and keep depth to 1–2.”