Live MCP session replay with real captured payloads

A/B: Teaching Graphonomous to Navigate a Real Codebase

Same user ask, two outcomes: A starts with no repo-specific procedural memory; B learns traversal skills from outcomes, links them to a goal, and answers architecture questions with clearer, faster traces. All payloads below are from a real Graphonomous MCP session.

Chat + tool-call storytelling

Live tool-call log

idle

Payloads captured from a real Graphonomous MCP session via capture_demo_tool_calls.py. Node IDs, confidence values, and latencies are authentic.