Skip to content

Chain of Thought

Layer 2: Thinking

Purpose

Forces reasoning to unfold in labeled stages so assumptions, decomposition, synthesis, and risks stay explicit
Helps downstream personas audit logic and reuse intermediate artifacts

When to Use

Architecture, planning, financial modeling, investigative analysis
Anytime problems benefit from multi-stage thinking or branching scenarios
When you need transparent reasoning for QA or regulated domains

Structure / Template

[STAGE 1: Assumptions]
List known facts, constraints, and data gaps.

[STAGE 2: Decomposition]
Break the problem into components, lenses, or sub-questions.

[STAGE 3: Exploration]
Evaluate options/paths for each component; note trade-offs.

[STAGE 4: Synthesis]
Integrate findings into a coherent recommendation or narrative.

[STAGE 5: Risks & Next Actions]
Flag uncertainties, edge cases, validations, and follow-up steps.

Examples

Technical (infra upgrade)

Assumptions: two-node Proxmox, shared ZFS, maintenance window 6h.
Decomposition: networking, storage, VM migration, rollback.
Exploration: compare live-migrate vs. cold-migrate, evaluate bandwidth.
Synthesis: choose cold-migrate with staggered reboot plan.
Risks & Next Actions: network congestion, take pre-upgrade snapshots.

Creative/strategic (market entry)

Assumptions: launch AI prompt wiki public beta, audience = DevOps + educators.
Decomposition: messaging, channels, onboarding content.
Exploration: evaluate blog, webinars, partner demos.
Synthesis: 3-stage rollout with educator pilot first.
Risks & Next Actions: measure retention, gather testimonials.

Combination Guidance

Pair With	When
Planning Phase	Turn CoT outputs into actionable plans
Rule-Based Reasoning	Each stage must cite explicit rules
Recursive Self-Eval	Critique individual stages before finalizing

Failure Modes to Avoid