Reasoning Topology

Reasoning topology is the shape of legible intermediate reasoning — not the domain conclusion, but how steps relate: linear, branching, merging, looping, parallel.

Chain of thought (CoT) is the linear topology. When reasoning is not a single line, name the topology honestly in capture (topology: tree | dag | network) — see Thinking by Writing for event types and broadcast.

Event log vs in-model prompting

External LLM research optimises prompting inside one model. Key references:

Taxonomy survey: Demystifying Chains, Trees, and Graphs of Thoughts — topology vs reasoning schedule.

Not standard names: Mesh of Thought, Loop of Thought, Hyper-Thought. Loops = feedback edges in GoT.

Event-sourced layer: publishing a thought as tier-one assert is broadcast — other agents react via causationId under one correlationId; converging, diverging, or recursive branches form a graph of thought events on the log. Pattern: Thought topology. Detail: Thinking by Writing — Reasoning topology.

Capture guidance

• CoT → ChainOfThoughtEmitted or chained ThoughtEmitted with causationId.
• Tree / graph → ThoughtGraphEmitted + envelope contextRef (MCP / harness).
• Pre-bind at automated binds: ReasoningAsserted.