DOT graphs as analysis infrastructure
Drawing a system as a DOT graph looks like harmless documentation. The dot-graph bundle argues it's something else entirely — and its own shipped materials say so out loud.
That claim isn't marketing. It's written into the code.
The bundle's dot-as-analysis skill states it directly — this is a released capability in the canonical Microsoft repo, at version 0.3.1.
So why would drawing a diagram surface anything new?
“DOT diagrams are not just documentation — they are analysis instruments.”
You believe a flow exists. You believe a box connects. Ordinary documentation lets both beliefs stand unchallenged — nothing forces you to check whether reality still agrees.
Drawing changes that. Here's the mechanism.
You cannot leave a box disconnected without asking why. You cannot draw a flow without tracing where it actually goes.
The bundle turns that pressure into a repeatable process.
“Completeness pressure is the mechanism. You cannot leave a box disconnected without asking why.”
The dot-as-analysis skill defines a named workflow. Each phase pushes belief a step closer to what the code actually does.
Run the last phase and the payoff appears.
In the skill's canonical teaching scenario, a “believed” payment-retry node turned out not to exist — exposing silently dropped transactions.
The diagram was the byproduct. The defect was the deliverable.
Believed: a payment-retry node.
Reality: it didn't exist.
Result: transactions dropped in silence.
The single dot_graph tool's analyze operation is NetworkX-backed, so reconciliation is verifiable rather than asserted.
And the whole thesis is operationalized end-to-end.
An agent authors it, an independent reviewer PASS/WARN/FAILs it, and an 11-agent Parallax pipeline triangulates the codebase from three perspectives.
The diagram is the byproduct. The reconciled understanding is the deliverable.
Data as of: Canonical repo last pushed 2026-05-18; latest commit & tags v0.3.0/v0.3.1 dated 2026-05-17.
Feature status: Shipped / released — microsoft/amplifier-bundle-dot-graph at version 0.3.1 (222 commits, 3 merged PRs). README notes external contributions are not currently accepted.
Primary source: Fresh clone of git@github.com:microsoft/amplifier-bundle-dot-graph.git. The claimed colombod/amplifier-bundle-dot-graph is a fork (isFork:true, parent owner microsoft); microsoft is canonical (isFork:false, parent:null).
Commands run:
gh repo view colombod/… --json isFork,parent & gh repo view microsoft/… --json isFork,parent — fork vs. canonicalhead -20 bundle.md — version 0.3.1; git tag — v0.3.0, v0.3.1 (both 2026-05-17)grep -n operation modules/tool-dot-graph/…/__init__.py — single dot_graph tool, 6 operations (validate, render, setup, analyze, prescan, assemble)grep -n 'import networkx…' analyze.py — analyses: stats, reachability, unreachable, cycles, paths, critical_pathsed -n '1,45p' skills/dot-as-analysis/SKILL.md — verbatim thesis + 4-phase workflow (Introspect → Represent → Reconcile → Surface)grep -n 'PASS|WARN|FAIL' agents/diagram-reviewer.md; ls agents/*.md | wc -l — 13 agents incl. 11-agent Parallax discovery pipelinegit log --format='%an' | sort | uniq -c | sort -rn — contributorsPrimary contributors: Brian Krabach (195 commits); Amplifier bot/automation (26); Salil Das (1).
Gaps: The payment-retry / dropped-transactions story is an illustrative teaching Scenario in the skill doc, not a verified production outage. No published GitHub Releases exist (git tags only). Analysis used a clone in /tmp; the local amplifier-module-tool-diagrams is an unrelated Kroki-based module and was excluded.