Mental Compound Interest: Cross-Domain Knowledge in Practice

Warren Buffett attributes his investment success to reading approximately 500 pages per day for over 60 years.¹ The compounding isn’t in the pages. It’s in the connections between them.

Knowledge compounds on a combinatorial dimension: each new concept creates potential connections with every existing concept, producing super-linear growth in useful insights. A 10x increase in concepts yields a 100x increase in pairwise connections. Cross-domain knowledge is where the highest-value connections emerge, because insights from one discipline unlock solutions invisible within a single field.

TL;DR

Knowledge compounds like financial capital, but on a combinatorial dimension: each new concept creates potential connections with every existing concept. After 12 years in product design, then transitioning to full-stack engineering and AI infrastructure, I’ve experienced this compounding firsthand. My blog quality system (a 12-module linter with 77 tests) exists because design principles, engineering patterns, and AI agent orchestration intersected in my head. No single discipline could have produced it. The interactive calculator below demonstrates the math behind knowledge compounding.

How Knowledge Compounds

The Connection Multiplier

Financial compound interest grows on a single dimension: principal increases by a fixed percentage. Knowledge compounds on a combinatorial dimension: each new concept creates potential connections with every existing concept.²

A knowledge base of 100 concepts contains approximately 4,950 potential pairwise connections (100 x 99 / 2). A knowledge base of 1,000 concepts contains approximately 499,500. A 10x increase in concepts produces a 100x increase in connections.

Not every connection is valuable. Most are irrelevant. But the subset of valuable connections grows super-linearly with knowledge base size — the challenge is identifying which connections matter, a problem I solved deterministically with a four-dimension signal scoring pipeline. Charlie Munger’s “latticework of mental models” describes exactly this: insights emerge from the intersection of models drawn from different disciplines.³

The Cross-Domain Advantage in Practice

Here’s where the math gets real. My blog evaluator combines:

From design (12 years): Visual hierarchy principles, readability scoring, content structure patterns from studying 16 products (Warp, Vercel, Linear, Raycast, Stripe, Figma, and others). These gave me the evaluation criteria.

From engineering (3 years of focused practice): Python regex, AST-like document parsing, test-driven development, CLI tooling. These gave me the implementation.

From AI agent work (1 year): Confidence scoring, multi-agent deliberation, consensus validation. These gave me the orchestration layer that connects linting to evaluation to fixing.

No single discipline could have built this tool. A designer would have written evaluation criteria but not the parser. An engineer would have built the parser but not the evaluation framework. An AI specialist would have built the orchestration but not the domain-specific quality rules.

The cross-domain connection produced a tool that none of the individual disciplines would have generated.⁴

The Reading Habit at Scale

The combinatorial explosion explains why Buffett’s reading habit is strategy, not dedication. Large knowledge bases are qualitatively different from small ones, not just quantitatively larger.⁵

My Compound Knowledge System

The Memory Layer

I maintain a structured memory system in .claude/MEMORY.md that captures errors, decisions, and patterns across sessions. This isn’t a journal. It’s a queryable knowledge base that grows with every project:

• Errors section captures failure patterns (spawn budget blocking agents, false positive quote detection, Bearer token quoting issues)
• Decisions section tracks architectural choices and their rationale
• Patterns section documents reusable approaches

Each entry connects to work from different domains. A “quoting false positive” error from bash hook development informed a regex improvement in the Python blog linter. A “CSRF blocking admin cache reload” decision from web development influenced the hook architecture’s approach to state mutation.

The connections happen because the memory system makes cross-domain patterns visible. Without it, lessons from one project stay trapped in that project’s context.

The Design Study Collection

I’ve studied 16 exceptional products in systematic detail: their interaction patterns, visual hierarchy decisions, performance strategies, and architectural choices. Each study adds to the lattice:

• Superhuman’s 100ms rule influenced my hook timeout thresholds
• Linear’s keyboard-first design informed my CLI interaction patterns
• Stripe’s documentation-as-product philosophy shaped my blog quality standards
• Perplexity’s citation-forward answers informed my footnote linting system

These connections aren’t planned. They emerge from having a large enough knowledge base that new problems remind me of old solutions from unrelated domains.

Mental Models as Investment Vehicles

Munger recommended building a “latticework” from the fundamental models of major disciplines:⁶

A practitioner who internalizes 3-5 models from each discipline operates with 20+ lenses for interpreting any new situation. I apply design hierarchy thinking to error messages, engineering redundancy thinking to quality gates, and economic opportunity cost thinking to deciding when to invest analysis.⁷

The Forgetting Curve Counter-Strategy

Hermann Ebbinghaus demonstrated in 1885 that humans forget approximately 70% of new information within 24 hours without review.⁸

Spaced Repetition

Systems like Anki reduce the forgetting rate from 70% to approximately 10% while requiring only 5-10 minutes daily. I built Ace Citizenship specifically around this principle: spaced repetition for USCIS civics questions.⁹

Writing as Consolidation

Writing about a concept forces active recall and structural organization, both of which strengthen memory formation more effectively than passive re-reading. My 29 blog posts serve this purpose: each post consolidates a knowledge domain into a queryable artifact.

The writing doesn’t need to be public. My .claude/MEMORY.md file serves the same consolidation function for operational knowledge. The act of writing is the investment. The artifact is a bonus.

Key Takeaways

For individual practitioners: - Start a daily reading habit across multiple disciplines, not just your primary field; cross-domain connections are rarer and more valuable than within-domain connections - Build a structured memory system (notes, wiki, markdown files) that makes cross-domain patterns visible; insights emerge from intersections you can’t see without a system - Use writing to consolidate learning; active recall during writing strengthens memory 3-5x versus passive re-reading

For managers building learning cultures: - Allocate 10-20% of work time to learning; the compound returns exceed the short-term productivity cost within 12-18 months - Create mechanisms for knowledge sharing across teams; cross-pollination produces the same combinatorial advantage that cross-disciplinary reading produces for individuals

FAQ

How does knowledge compound differently from financial capital?

Financial compound interest grows on a single dimension: principal increases by a fixed percentage. Knowledge compounds on a combinatorial dimension: each new concept creates potential connections with every existing concept. A knowledge base of 100 concepts has approximately 4,950 pairwise connections, while 1,000 concepts has approximately 499,500. A 10x increase in concepts produces a 100x increase in potential connections.

What is the cross-domain advantage?

The cross-domain advantage occurs when knowledge from unrelated disciplines intersects to produce tools or insights that no single discipline could generate alone. The author’s blog quality system combines design evaluation criteria (12 years of design experience), Python engineering implementation (3 years of focused practice), and AI agent orchestration (1 year of agent work). A designer, engineer, or AI specialist working alone could not have built the same tool.

How do I build a system for capturing cross-domain connections?

Maintain a structured memory system (notes, wiki, or markdown files) organized by errors, decisions, and patterns rather than by project or domain. The author uses .claude/MEMORY.md as a queryable knowledge base. When lessons are recorded by category rather than project, connections between domains become visible. A quoting false-positive error from bash hook development informed a regex improvement in a Python blog linter because both were recorded as pattern entries.

What is the forgetting curve and how do I counter it?

Hermann Ebbinghaus demonstrated that humans forget approximately 70% of new information within 24 hours without review. Two proven counter-strategies: spaced repetition systems (like Anki or Ace Citizenship) reduce the forgetting rate to approximately 10% with 5-10 minutes of daily review, and writing about concepts forces active recall and structural organization that strengthens memory 3-5x versus passive re-reading.

References