No labels. No embeddings. No gradients. Watch structure emerge from time order alone.
ScenarioSpeed8 tok/s
The rule
When A fires, then B fires shortly after, the
A→B edge gets stronger. That's the entire learning rule -
no labels, no gradient descent, no global counts.
waiting for first token...
Input
sequence -
of -:
(scenario loading...)
step 0 / 0
Live graph — what murmur built so far- edges · - concepts
▬ matches textbook
▬ extra (longer-window correlation)
Expected structure — the textbook answer-
Same node positions in both panels. Each violet edge
on the right is a transition the textbook expects to dominate.
Compare directly: does the live panel have a corresponding dark edge?
about this scenario ▸
-
-
Output — ask the graph:
-
Seed
Resultspick a seed
concept
random
baseline
murmur
no seed selected.
View
Edge graph: nodes pinned to textbook layout, dark arrows = learned edges.
Layers (overlay)
Internal grouping the engine forms automatically. Off by default; advanced viewers only.