liveBook · Manning

Chain-of-thought

As explained in The Architecture of Reasoning

A reasoning technique that proceeds through ordered analytical steps.

A payment service starts timing out under load. The team blames the database and prepares to increase its capacity. But the visible sequence tells a different story: a reconciliation job consumes CPU, latency rises, retries amplify the load, and the database suffers downstream. The quick answer named a symptom; the intermediate steps exposed a different cause. section 3.1 section 3.3.1

Multi-step reasoning, or chain-of-thought, makes the path from observation to conclusion visible before a decision is accepted. Instead of debating finished recommendations, a team can inspect its inferences, assumptions, and causal links. The aim is not to narrate every thought. Show the steps that materially affect the conclusion, especially the transitions that might fail. section 3.2 section 3.3

A recommendation with its assumptions exposed

“We should enter this market because demand is growing” sounds complete. Unpacking it reveals observations about adjacent-market growth and acquisition costs, an inference that existing capabilities will transfer, and a judgment that entry risk is acceptable. The team can now ask whether those signals predict this market, whether cost trends will persist, and whether the implied risk fits its tolerance—without treating the recommendation as wholly right or wrong. section 3.3.1

Use the extra structure when compression is risky:

  • The problem is unclear, and people may be answering different interpretations of it. section 3.5.1
  • An error would be expensive to reverse or its effects would emerge slowly. section 3.5.2
  • Participants bring different domain assumptions, incentives, or definitions of success. section 3.5.3
Table 3.1 A compact reasoning scaffold for chain-of-thought
Element
Purpose
Guiding question
Problem definition
Establish a shared frame
What exactly are we trying to understand or decide?
Observations
Ground reasoning in facts and signals
What do we know to be true?
Inferences
Make interpretation explicit
What are we inferring from this?
Assumptions
Surface dependencies
What must be true for this to hold?
Causal chain
Test logical connections
How does each step lead to the next?
Conclusion
Make the decision explicit
Given this, what should we do?
Fragility check
Expose weak points
Where is this reasoning most vulnerable?

A compact scaffold moves from a shared problem definition through observations, inferences, assumptions, and causal links to a conclusion and a fragility check. Use it aloud or in a few lines; it need not become formal documentation. section 3.7

Visible steps improve inspectability, not certainty: a false assumption can still support an orderly chain. The practice also fails when questioning is treated as a personal challenge, only some people must explain themselves, or every small decision becomes a lengthy exercise. For a bounded, easily corrected question, Direct Q&A may be enough. When the stakes justify a closer look, ask: “What happened between that signal and this conclusion?” Then examine the weakest link before committing. section 3.2 section 3.7 section 3.8 section 7.2

How The Architecture of Reasoning covers it

How deep it goes
Develops a conversational sequence and compact checklist for examining assumptions and causal links, without requiring formal documentation.
Examples
Uses a payment-service incident and a market-expansion recommendation to reveal hidden causal leaps and assumptions.
What it assumes
Assumes familiarity with workplace decisions and introduces the stepwise reasoning pattern through contrasts with direct answers.
What you can do after
Practice/perform stepwise questioning to locate weak links before an ambiguous or costly decision is made.
How it connects
Contrasts with Direct Q&A and connects to Tree-of-Thought, Chair-of-Thought, and ReAct as selectable reasoning structures.

Also explained in

Compare how 21 books explain chain-of-thought →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of The Architecture of Reasoning. The section links point to the parts of the book it draws on.