Skip to content

methodology.md

Latest commit

 

History

History
76 lines (49 loc) · 3.87 KB

methodology.md

File metadata and controls

76 lines (49 loc) · 3.87 KB

Meta-Synthesis Methodology Background

综合集成法方法论背景

Origin 起源

Meta-Synthesis (综合集成法) was proposed by Qian Xuesen (钱学森) in the late 1980s as an approach to solving complex, open systems problems — what he called "open complex giant systems" (开放的复杂巨系统).

The core insight: complex systems cannot be understood by any single discipline or method alone. You need the synthesis of:

  1. Qualitative knowledge from domain experts
  2. Quantitative data from measurements and computation
  3. Systematic integration that finds convergence across both

Hall for Workshop of Meta-Synthetic Engineering (HWMSE)

综合集成研讨厅

Qian proposed HWMSE as a practical framework with three interacting subsystems:

Expert System 专家体系

Human experts contribute domain knowledge, intuition, experience, and judgment. They identify patterns that machines miss and recognize when quantitative results "don't feel right."

Knowledge System 知识体系

The accumulated body of documented knowledge: theories, models, specifications, past decisions, and recorded lessons. This system provides consistency checks and institutional memory.

Machine System 机器体系

Computational tools that process data, run simulations, verify hypotheses, and produce quantitative results. Machines are fast and precise but lack contextual judgment.

The Integration Principle 集成原则

No single subsystem has veto power, but all three must converge.

When they agree, confidence is high. When they disagree, that disagreement itself is the most valuable signal — it reveals hidden assumptions, missing data, or flawed models.

From System Engineering to Software Engineering

The Problem: Cognitive-Reality Gap in AI-Assisted Development

In vibe coding (AI-assisted development), a specific failure mode emerges:

  • The developer sees Claude generate clean, plausible code
  • The code reads well (Expert subsystem: "looks good")
  • But it hasn't been validated against specs (Knowledge) or tested by machines (Machine)
  • The developer ships with false confidence

This is a convergence failure — only one of three subsystems has approved.

The Mapping

HWMSE Subsystem Software Engineering Context
Expert System Claude's analysis + developer's review and judgment
Knowledge System CLAUDE.md, memory files, ADRs, specs, design docs
Machine System Linters, type checkers, test suites, build tools, security scanners

The Iron Law

NO COMPLETION WITHOUT THREE-SYSTEM CONVERGENCE

This is the software engineering translation of Qian's integration principle. It operationalizes the insight that convergence across qualitatively different validation methods produces higher confidence than any single method alone.

Key Principles Adapted for Software

  1. Qualitative before quantitative: Understand the problem space (Phase 1) before writing code (Phase 2).

  2. Small increments with verification: Each code increment is validated by the Machine subsystem immediately, not batched.

  3. Convergence over coverage: 100% test coverage with Expert dissent is worse than 80% coverage with all three systems aligned.

  4. Dissent is signal, not noise: When the Expert feels uneasy but tests pass, the unease is valuable data. Investigate.

  5. Knowledge maintenance is not optional: Lessons learned must be captured (Phase 5) or they are lost to temporal decay.

References

  • 钱学森, 于景元, 戴汝为. "一个科学新领域——开放的复杂巨系统及其方法论", 自然杂志, 1990.
  • Qian, X.S., Yu, J.Y., Dai, R.W. "A New Discipline of Science — The Study of Open Complex Giant Systems and its Methodology", Chinese Journal of Nature, 1990.
  • 戴汝为, 李耀东. "综合集成方法与系统复杂性", 系统工程理论与实践, 2004.