{"schemaVersion":"1.0","type":"Article","title":"My Thinking Method: Go Upstairs, Compress the Structure, Then Return Downstairs","description":"Start from anomalies, find underlying variables and cross-domain structures, compress them into a model, and return to action—while using falsification, boundaries, and evidence to control elegant theories.","author":{"name":"Zhao Bo","alternateName":"赵波","profile":"https://xinjignxiaozhaobo.com/en/about/"},"publisher":"Zhao Bo (赵波)","language":"en","publishedAt":"2026-08-27T00:00:00.000Z","updatedAt":"2026-08-27T00:00:00.000Z","topic":{"name":"Life & Thinking","url":"https://xinjignxiaozhaobo.com/en/topics/life-thinking/"},"tags":["Thinking models","Cognitive compression","Cross-domain transfer","Falsification"],"translationKey":"upstairs-thinking-model","canonical":"https://xinjignxiaozhaobo.com/en/upstairs-thinking-model/","markdown":"https://xinjignxiaozhaobo.com/en/upstairs-thinking-model.md","json":"https://xinjignxiaozhaobo.com/api/articles/en/upstairs-thinking-model.json","translation":{"language":"zh-CN","canonical":"https://xinjignxiaozhaobo.com/zh/upstairs-thinking-model/","markdown":"https://xinjignxiaozhaobo.com/zh/upstairs-thinking-model.md"},"citation":"Zhao Bo. “My Thinking Method: Go Upstairs, Compress the Structure, Then Return Downstairs.” 2026-08-27. https://xinjignxiaozhaobo.com/en/upstairs-thinking-model/","copyright":"Copyright © 2026 Zhao Bo (赵波)","usagePolicy":"https://xinjignxiaozhaobo.com/ai-policy.txt","contentFormat":"text/markdown","content":"Looking back at my work across consumption, brands, AI, organizations, retail, and technological history, I see that what I repeatedly use is not one body of knowledge but a relatively stable sequence of cognitive moves.\n\nI currently call it:\n\n> **Upstairs observation — structural compression — cross-domain transfer — downstairs reconstruction**\n\nIt works roughly like this:\n\n> **Phenomenon → anomaly or tension → step outside the phenomenon → find underlying variables → build a structural model → look for isomorphism across domains → compress it into a higher-order explanation → return to the specific problem → turn it into a product or action**\n\nThe point is not to make an idea sound deep. It is to explain more phenomena with fewer core variables and ultimately improve decisions in the real world.\n\n## The starting point is not an answer but a sense that something is wrong\n\nMany of my inquiries begin not with a clear question but with a feeling:\n\n> Everyone explains it this way, but the explanation seems incomplete.\n\nFor example, saying that consumption is changing because “young people want to please themselves” describes a surface pattern. It does not explain why the pattern appeared. Keep asking, and a longer chain emerges: more choice, weaker external authority, stronger agency, a transfer in the right to define the self, a new ranking of consumer value, and only then a change in brand narrative.\n\n“Self-pleasing consumption” stops being an explanation and becomes an outcome that itself requires explanation.\n\nThe key move is to remain dissatisfied with second-order explanations. Instead of accepting the vocabulary already circulating in an industry, ask:\n\n> What generative mechanism produces this phenomenon?\n\n## Go upstairs: move from the object to the system\n\nMost problems first appear as concrete objects: how a store chooses products, what an agent can do, how a brand becomes younger, or why a transformation fails.\n\nGoing upstairs does not mean abandoning reality. It means temporarily stepping away from the object and rewriting the question as a system:\n\n- Where does information enter?\n- Who interprets it?\n- Who acts?\n- How do resources move?\n- What are the constraints?\n- How does feedback return?\n- Which link determines the outcome of the system?\n\nWhen thinking about assortment, I stop looking only at products and examine how environmental information, store state, consumer demand, product decisions, and feedback form a loop. When thinking about agents, I do not simply list features. I reconstruct the relationships among people, agents, skills, knowledge, data, permissions, and governance.\n\nThis is the move from object thinking to system thinking: leaving the pieces on the board for a moment and looking at how the whole game operates.\n\n## Look for structural isomorphism across domains\n\nAfter abstracting a problem, I often ask another question:\n\n> Does this structure exist in a completely different field?\n\nConsumer agency, the psychological idea of a persona, modern brand narrative, and the third era of consumption belong to different domains. Yet they may share one relation: movement from “others define me” toward “I participate in defining myself.”\n\nCompanies, nervous systems, and AI agents should not be treated as equivalent. Yet each may deal with information input, state assessment, action selection, feedback, and survival constraints.\n\nThis is not random association. It is a search for structural isomorphism: different objects, but perhaps a similar pattern of relationships.\n\nThe benefit is transfer. A question that has proved useful in one field can illuminate a hidden variable in another. But one rule is essential: **similarity is not identity of mechanism**. Analogy can generate a hypothesis; it cannot substitute for evidence.\n\n## Compress: theory reduces the cost of explanation\n\nAs the number of variables grows, I try to compress them into a chain, several layers, a coordinate system, a diagram, or one central proposition.\n\nIf dozens of phenomena cannot be compressed into a small number of key relationships, I usually assume I do not yet understand them.\n\nA useful model can be described with a simple ratio:\n\n> **Model value = range of phenomena explained ÷ complexity of the model itself**\n\nCompression is not an attempt to pretend the world is simple. It is a way to discover which variables carry explanatory power. A useful model should make prediction, comparison, and action easier—not merely memorization.\n\n## Return downstairs and let reality test the model\n\nAbstraction is not the destination.\n\nWhenever a theory begins to take shape, I eventually have to return downstairs and ask:\n\n- How should the product change?\n- Which process must be redesigned?\n- What should the brand say, and what should it stop saying?\n- Which data could test this?\n- What can be done tomorrow?\n- If the model is wrong, which signal will reality produce?\n\nThe complete path is not simply practice to theory. It is:\n\n> **Practice → abstraction → theory → practice again**\n\nOnly on returning to the concrete problem does a model reveal its boundaries. A theory that cannot change observation, choice, or action may be no more than an elegant linguistic structure.\n\n## An eight-step algorithm that can be reused\n\nThe method can be organized into eight repeatable steps.\n\n### 1. Capture the anomaly\n\nWhere does the prevailing explanation feel incomplete? Which fact does not fit the dominant story?\n\n### 2. Remove the labels\n\nTemporarily put aside industry phrases such as “emotional value,” “digital transformation,” or “AI empowerment.” Describe what is actually happening.\n\n### 3. Find the generative mechanism\n\nWhich variable had to change for the phenomenon to appear? Was it capability, cost, power, media, incentives, or constraints?\n\n### 4. Go upstairs\n\nTranslate the concrete object into systems, relationships, roles, information, resources, feedback, and constraints.\n\n### 5. Search for cross-domain isomorphism\n\nDoes history, nature, technology, organizational life, psychology, or another industry contain a process with a similar structure?\n\n### 6. Compress it into a parent model\n\nUse a chain, hierarchy, coordinate system, flywheel, or central proposition to express the small number of variables that truly matter.\n\n### 7. Expand again\n\nIf the model is true, what else should it explain? What new predictions does it generate?\n\n### 8. Return downstairs\n\nWhat should the product do, how should the organization change, what data can test the claim, and what happens next?\n\n## Why the thinking can sometimes appear to jump\n\nHigh compression creates a communication problem.\n\nThe reasoning in my head may run A → B → C → D → E → F. Because B, C, and D are already familiar to me, what I say aloud becomes A → F. It feels natural internally, but the bridge is invisible to everyone else.\n\nThe logic is not necessarily absent. Its compression ratio is too high for the available communication bandwidth.\n\nThe solution is not to abandon abstraction. It is to restore three kinds of information deliberately:\n\n1. The intermediate evidence between fact and judgment;\n2. Exactly which relationships are preserved when an idea moves from one domain to another;\n3. The conditions, boundaries, and exceptions under which the theory applies.\n\nPowerful expression does more than deliver a conclusion. It allows another person to follow the same path and audit the result.\n\n## Three dangers of elegant grand models\n\nThe greatest risk of this method is falling in love with a beautiful model that seems able to explain everything.\n\n### Over-unification\n\nOne variable may explain A, B, and C without driving the entire world. Reality is often produced by several mechanisms operating together.\n\n### Analogy contamination\n\nA company may resemble an operating system, an agent may resemble an employee, and a brand may resemble a personality. These analogies can be generative, but they do not prove identical mechanisms. We must state where the structures match and where the similarity is only rhetorical.\n\n### Model production outruns model consolidation\n\nContinuous invention produces many concepts with partial value but no reusable system. The more important task is to compress local models into a small number of parent models and expose them to repeated tests.\n\nFor every elegant theory, I should therefore force myself to answer three questions:\n\n> **Which phenomenon can this theory not explain?**\n\n> **What evidence would make me admit that it is wrong?**\n\n> **Is there another model that explains the same phenomenon?**\n\nFalsification, boundaries, and evidence are not enemies of abstraction. They are the conditions that make abstraction reliable.\n\n## Why model generation matters more in the AI era\n\nAI is lowering the cost of search, writing, analysis, coding, and execution. Scarcity therefore moves upward: Which problem is worth solving? What is the problem really? Which model should we use? Can two previously disconnected fields illuminate one another? Can a new explanatory frame lead to action?\n\nI prefer to describe my core capability as model generation: finding hidden structure in a mass of phenomena, compressing it into a model, and transferring that model into a new problem.\n\nBut the value of that capability must be multiplied by the strength of validation and reduced by the losses caused by unexplained jumps.\n\nThe full method therefore needs more than the four moves of abstraction, compression, transfer, and reconstruction. It also needs falsification, boundaries, evidence, and expression.\n\nGoing upstairs helps us see the structure. Compression helps us identify what matters. Returning downstairs allows reality to decide whether the model deserves to survive."}