DOT Graph Intelligence

Overview

Structural questions about graphs — "is this node reachable?", "are there cycles?", "what changed?" — are code problems, not language model problems. Graph analysis tools answer these in milliseconds with zero ambiguity. Reserve LLM reasoning for interpretation, not computation.

Core principle: Code answers structural questions — that is code's judgment to make. LLMs interpret what the answers mean. Use the right tool for each layer.

---

When to Use

digraph when_to_use {
    label="When to Use Graph Intelligence"
    labelloc=t
    rankdir=TB
    node [shape=diamond style="rounded,filled" fillcolor="#FFF9C4" fontname="Helvetica"]
    edge [fontname="Helvetica" fontsize=10]

    Q1 [label="Is the question\nstructural?\n(reachability, cycles,\npaths, counts)"]
    Q2 [label="Does it require\nunderstanding\nmeaning or context?"]

    node [shape=box style="rounded,filled"]
    UseCode [label="Use graph analysis\noperations" fillcolor="#C8E6C9"]
    UseLLM  [label="Use LLM reasoning\nwith graph as context" fillcolor="#E8F0FE"]
    UseBoth [label="Run analysis first,\nthen interpret results\nwith LLM" fillcolor="#FFF9C4"]

    Q1 -> UseCode [label="yes — pure structure"]
    Q1 -> Q2      [label="no"]
    Q2 -> UseLLM  [label="yes — meaning only"]
    Q2 -> UseBoth [label="both structure\nand meaning"]
}

---

Operations Reference

Operation	Question It Answers	Input	Output
reachability	Can node A reach node B?	source node, target node	yes/no + path
unreachable	Which nodes are isolated or stranded?	graph	list of unreachable nodes
cycles	Are there circular dependencies?