Serendipity

Grounded serendipity

graph_discover_grounded({ nodes: 3, intensity: 0.2 }) — random nodes with a real bridge.

Two-phase protocol:

1. System provides random nodes → you find the real bridge (shared substrate or functional analogy)
2. System injects noise into your bridge → you rationalize the corruption into new territory

Phase 1: Sense-Making — find the real bridge before inventing. Look for structural analogies, shared mechanisms, or conceptual overlaps between the random nodes.

Phase 2: Chaos Perturbation — the system corrupts your bridge statement. Your job is to treat the corruption as an axiom and invent new theory from it.

Use grounded serendipity for lasting theory. The result should be a serendipity or surprise node connected to the source nodes.

Pure serendipity

graph_serendipity() — high novelty, lower coherence.

The serendipity engine forces unexpected connections through enforced blindness:

• graph_discover returns ONLY a prompt — the blind agent doesn't see source nodes
• The blind agent treats corrupted seeds as axioms and invents physics
• Cold nodes (rarely accessed) are prioritized for unexpected combinations

Use pure serendipity for breaking blocks.

When to use

• Thermostat says DIVERGE
• Graph density is high (tight mental model needs disruption)
• You're stuck in a rut
• The same triggers keep appearing (too much foundation and decision)

Follow through

Divergence is not decoration. If graph_discover_grounded returns a genuine bridge, the next batch must write a serendipity or surprise node about it. A divergent tool call with no follow-up commit is looking at shiny objects.