Self-Learning First Principles Analysis

Problem

First-principles analysis is useful for financial valuation, industry research, and opportunity assessment, but it often collapses into pattern matching or conclusion-first reasoning: treating high growth as enough, rejecting high goodwill reflexively, or extrapolating industry cycles without checking assumptions. This skill turns the analysis into an inspectable workflow that starts from base facts, rebuilds a judgment, and records reusable conclusions.

How It Works

• Eight-step flow: from Step 0 memory loading to Step 1 decomposition, Step 2 Socratic questioning, Step 4 solution generation, Step 5 cross-validation, Step 6 knowledge distillation, and Step 7 reflection.
• Three analysis tracks: financial valuation uses L1-L7 layers covering earnings quality, asset quality, cash flow, DCF, relative valuation, SOTP, and real options; industry analysis follows “industry essence, value chain, competitive structure, drivers, and boundary inflection”; opportunity or difficulty cases use four sources, four types, and a quadrant map.
• Data and computation: prefer real market data, three financial statements, fundamentals, and fund-flow inputs; when data is missing, use deconstruct for logic-only decomposition, mark assumptions, and run sensitivity checks.
• Memory system: persist findings in knowledge-base.md, patterns.md, and reflections.md, promoting knowledge from “unverified” to “candidate” and then “confirmed.”

Limits

It fits professional analyses that need an explicit reasoning chain, assumption list, and confidence assessment. It should not be treated as a trading decision oracle. When input data is sparse, conclusions depend more heavily on assumptions, so pay attention to Step 5 sensitivity analysis and stated risk boundaries.