Exploring What You Built
With 56 nodes and 66 edges, snowflex's decision graph tells the complete story of how the library evolved. The real power isn't in reading nodes one by one — it's in asking your AI questions and getting answers grounded in the graph.
Asking Questions About Your Graph
The decision graph is not designed for humans to read directly. It's a database that your AI assistant queries when you have questions. You ask in plain language; Claude searches the graph and answers with evidence.
34 revisit active Reconsider ODBC as transport - switch to HTTP REST API
46 revisit active Reconsider single-key auth - support multiple key formats
Title: Reconsider ODBC as transport - switch to HTTP REST API
Incoming: #33 — "ODBC limitations forced rethinking"
Outgoing: #36 — "Revisit leads to new transport decision"
Title: ODBC accumulating problems - sessions expire after 4 hours, UTF-16 corruption, Apple Silicon incompatible, memory issues
Notice what happened: you asked a question in English, and Claude traced the full decision chain — from the observation, through the revisit, through the option evaluation, to the outcome. The graph gave it the evidence; Claude gave you the story.
More Examples
2 option superseded Use Erlang ODBC driver to connect to Snowflake
3 decision superseded How to connect Elixir to Snowflake?
5 outcome superseded Working single-connection ODBC wrapper
7 option superseded Explicit ConnectionPool in supervision tree
9 outcome superseded Multi-connection support via ConnectionPool (v0.1)
Once the graph has history, Claude uses it to avoid repeating past mistakes. If you ask it to implement something that was already tried and rejected, it will check the graph first and warn you — explaining why that approach failed before and what replaced it.
The Web Viewer (Optional)
For a visual overview, the built-in web viewer renders the graph in a browser. It's useful for presentations and getting a high-level sense of the graph's shape, but the Q&A interface above is how you'll interact with it day-to-day.
The viewer provides DAG, timeline, chains, and archaeology views. Leave it running in a background tab — it auto-refreshes as Claude adds nodes during a session.
Pulse — Graph Health Report
The deciduous pulse command gives a quick summary of your graph's health and structure:
Pulse is useful for catching problems: orphan nodes (nodes with no connections), dangling outcomes (outcomes not linked to an action), and overall balance of node types.