serendipity

Serendipity

Serendipity is an encounter with material outside the most likely completion path, followed by judgment. It is not a novelty quota, a claim that unrelated things must connect, or permission to abandon the user's task.

Choose the encounter

If the user has not requested a particular method and there is a genuine exploratory choice, start with:

graph_suggest_next({
  task: "State the live task or uncertainty",
  workflow: "general",
  creativity: true
})

Replace general with the live work domain when reading, research, coding, collaborative coding, or writing applies.

Consider higher-weighted routes seriously, then choose, combine, modify, or reject them. The model decides whether a proposed encounter can serve the live task; the menu is neither exhaustive nor mandatory. Do not call for another roll merely because one turn elapsed.

Pass creativity: false when ordinary guidance is useful but creative provocations are not. This affects that roll only; it does not disable any direct tool. Use graph_thermostat() when an entropy-style structural pulse would genuinely help decide between convergence, continuation, and divergence. Its pressure and recommended tool are advisory, not a command or quality score.

Choose the creative pressure

Use graph_bisociate to surface cross-context candidates through spreading activation and information gain. It is often the lightest way to leave a local groove: inspect a candidate's actual nodes, then decide whether a relation is real.

Use graph_discover_grounded({ nodes: 3, intensity: 0.2 }) when you want the graph to sample distant material and ask whether a real bridge exists. Inspect the actual sources, then look for a shared mechanism, constraint, structure, or functional analogy. A defensible no-connection result needs no node.

After finding a genuine bridge, graph_discover_grounded_chaos can perturb that bridge. Do not call it merely because the grounded tool returned nodes. Use graph_discover({ nodes: 2, cold: true, blind: true }) for the more speculative blind ANI path, or graph_chaos({ text, blind: true }) to perturb chosen text. Blind ANI is meaningful only when a separate model context sees the returned prompt without the original source context.