This viewer uses a fixed set of Game of Thrones characters as nodes and replays the edges emitted by a language model one step at a time. The ring order follows the prompt order, while the edge sequence follows the raw order produced by the selected model run.
How to read the overlay
Soft gray edges are the ground-truth graph. Green edges are emitted edges that match ground truth. Red edges are hallucinated edges that the model produced but that are not in the selected ground-truth network.
What the controls change
Prompt order switches between the default and reverse prompt variants. Model and Run choose a specific LLM output trace. Speed, Pause, Step, and Reset control the sequential playback.
What this page is for
The page is meant to compare how different models build the same graph over time, not just whether they end at the right answer. It highlights ordering effects, early mistakes, repeated structural patterns, and where hallucinated edges begin to accumulate.