Salesforce impact analysis

The problem with guessing

A general AI model will list "typical" references - validation rules, page layouts, reports - but cannot know which ones exist in your org. sf-intelligence builds a DuckDB dependency graph from a read-only metadata retrieve and traces edges with confidence tags: declared (from Salesforce metadata), parsed (from AST/XML analysis), or heuristic (regex - spot-check these).

Questions you can ask

• "What breaks if I delete the Contact email field?"
• "What is the blast radius of removing the Payment custom object?"
• "Is it safe to deactivate this Flow?"
• "What depends on Account.Industry__c?"
• "What would happen if I make this field required?" (optionally with live null-rate)

The router picks tools like find_component_usages, blast_radius, what_if_delete, and field_change_advisor automatically - you type the question, not the tool name.

Evidence tiers you can trust

Signal	Meaning
declared	Salesforce metadata states the dependency directly (highest trust)
parsed	Found by AST/XML parsing of Apex, Flow, or formula source
heuristic	Regex/token match - may have false positives; verify before acting
partial	A metadata family was not retrieved - "not checked", not "none"

Optional: live record counts

With the opt-in read-only live plane, blast_radius_live adds affected-record counts on top of the static graph. Hybrid answers disclose both vault freshness and live provenance - if the org is ahead of the vault, you get a staleness warning first.

Honest boundaries

sf-intelligence performs static analysis. Dynamic SOQL, reflective Apex, and runtime-only references are invisible. "No references found" means no static evidence, not "definitely unused." See the full boundary list on the trust page.