You have a factory you have never inspected
You build mental models constantly. Every time you start a new job, meet a new person, learn a new tool, enter an unfamiliar situation — you construct a schema. A working theory of how this thing operates, what matters, what to expect. You have been doing this since childhood. Thousands of schemas, built over decades.
But here is the question L-0321 opened up and that most people never ask: how do you build them? Not what your schemas contain — how they get constructed in the first place. What is the process? What raw materials do you reach for? What sequence do you follow? What do you skip?
You have a schema creation process. It runs every time you need a new mental model. And unless you have deliberately examined it, that process is invisible to you — which means every model it produces inherits the same unexamined biases, the same structural blind spots, the same characteristic weaknesses.
Knowing the content of your schemas is Phase 16 work. Knowing how those schemas get made — that is Phase 17 work. That is the meta-schema that governs all the others.
How schemas actually form: the cognitive science
Schema formation is not a single mechanism. It is a layered process that cognitive science has mapped from multiple angles, each revealing a different part of the machinery.
Piaget's constructivism provides the foundational frame. Jean Piaget demonstrated that knowledge is not passively received — it is actively constructed through interaction with the environment. His two core mechanisms are assimilation (integrating new information into an existing schema without changing the schema's structure) and accommodation (modifying or creating a schema when new information cannot fit the existing one). The interplay between these two processes — what Piaget called equilibration — is the engine of cognitive development. You encounter something new. You try to fit it into what you already know. When it does not fit, the resulting disequilibrium forces a structural update. That update is a new or revised schema.
But Piaget's model describes the trigger for schema change more than the construction process itself. How does the new schema actually get built?
Bartlett's schema theory (1932) showed that schemas are constructed from memory, not from raw perception. When you form a new mental model, you are not working from a blank slate — you are assembling it from fragments of prior experience, cultural defaults, and expectations. Bartlett demonstrated this with his famous "War of the Ghosts" study: participants who read an unfamiliar Native American folktale did not remember it accurately. They reconstructed it to fit their existing cultural schemas, smoothing out the unfamiliar elements and inserting familiar ones. Schema creation, in other words, is always schema re-creation. You build new models from the parts of old ones.
The chunking research from chess expertise adds another layer. Adriaan de Groot (1965) discovered that chess masters could reconstruct board positions after a five-second glance — but only for positions from real games. Random positions eliminated the advantage entirely. Chase and Simon (1973) explained this through chunking: experts do not see individual pieces. They see patterns — clusters of pieces that form meaningful configurations. Over thousands of hours, these chunks accumulate into a vast library of schemas that allow the expert to recognize a new position by matching it against stored patterns.
This finding generalizes far beyond chess. Whether you are a doctor reading a scan, a programmer reading code, a manager reading a room — your expertise is your chunk library. Your schema creation process is the process by which new chunks get formed and integrated.
The five common schema creation modes
When you watch how people actually form new mental models — not how they think they do, but what they actually do — five distinct modes emerge. Most people default to one or two.
Mode 1: Inductive accumulation. You collect concrete experiences and gradually notice what they share. After enough encounters with a pattern, a schema crystallizes. This is the empiricist mode — bottom-up, data-driven, grounded in direct observation. Its strength is that the resulting schema is tightly coupled to reality. Its weakness is that it is slow and limited to patterns you happen to encounter personally.
Mode 2: Deductive adoption. You encounter a framework — in a book, a lecture, a conversation with someone you respect — and adopt it as a working model. This is the theorist mode — top-down, framework-first, grounded in authority or logical coherence. Its strength is speed and access to patterns beyond your personal experience. Its weakness is that adopted schemas may not match your actual context, and you may never test them against your own data.
Mode 3: Analogical transfer. You encounter something unfamiliar and immediately map it to something you already understand. "This new market is like the last one, except..." "Managing this team is like coaching a sports team because..." This is the metaphor mode — lateral, pattern-matching, grounded in structural similarity. Its strength is rapid schema construction for genuinely novel domains. Its weakness is that the analogy may hide critical differences between the source and target domains.
Mode 4: Social absorption. You adopt the schemas of the group you belong to without a conscious construction process at all. You believe what your industry believes, value what your peer group values, frame problems the way your organization frames them. This is the cultural mode — ambient, osmotic, grounded in belonging. Its strength is alignment and social coherence. Its weakness is that you may never hold an original thought in the domain — only inherited ones.
Mode 5: Experimental construction. You form a hypothesis, design a test, run it, and revise based on results. This is the scientific mode — iterative, falsification-oriented, grounded in deliberate testing. Its strength is that the resulting schema has been stress-tested against reality. Its weakness is that it requires time, discipline, and the willingness to be wrong — which is why it is the least common default mode.
Most people use Mode 1 and Mode 4 almost exclusively. They learn from personal experience and from the people around them. Modes 2 and 3 show up in people who read widely or think in systems. Mode 5 is rare outside of professional scientists and engineers — and even they often abandon it when the domain shifts from their technical work to their personal beliefs.
The point is not that one mode is correct. The point is that your default mode shapes every schema you build. If you always use Mode 1, every model you hold is limited to your personal sample size. If you always use Mode 2, every model is borrowed and potentially untested. Knowing your default is the first step toward expanding your repertoire.
Deliberate practice and the expert's schema factory
Anders Ericsson's research on deliberate practice reveals something crucial about schema creation: experts do not just have more schemas than novices. They have a fundamentally different schema creation process.
Ericsson showed that expert performance requires what he called increasingly complex mental representations — internal models that allow the expert to monitor, evaluate, and control their own performance in real time. These representations are not acquired through simple repetition. They are built through deliberate practice: structured training focused on specific weaknesses, with immediate feedback, and repeated attempts at the same task with targeted adjustments.
The key insight for schema creation is this: experts counteract automaticity. When novices practice, they reach a point where performance becomes automatic — good enough, habitual, no longer requiring conscious attention. Their schema creation process freezes. Experts, by contrast, deliberately stay in the cognitive and associative phases where schemas are still being actively constructed and refined. They resist the pull of "good enough."
Donald Schon's work on reflective practice describes the same dynamic from a different angle. Schon distinguished between knowing-in-action (the tacit knowledge you apply instinctively), reflection-in-action (adjusting your approach in real time while doing the work), and reflection-on-action (retrospectively examining what happened and why). The transition from knowing-in-action to reflection-on-action is the transition from running your schema creation process unconsciously to examining it consciously.
Most professionals operate almost entirely in knowing-in-action mode. They build schemas automatically, from experience, without examining the construction process. Reflective practitioners do something different — when they encounter a unique case that does not fit their existing models, they do not just force-fit it into existing schemas. They construct a new theory of the unique case. They treat schema creation as a craft to be practiced, not a reflex to be left alone.
The practical implication: the quality of your schemas is not determined primarily by your intelligence or your experience. It is determined by whether your schema creation process itself has been deliberately developed or left on autopilot.
Your AI as a schema construction partner
The parallel between human schema creation and how artificial neural networks learn representations is more than metaphorical — it reveals structural insights about the construction process itself.
A neural network does not start with schemas. It starts with random weights — no structure, no patterns, no models. Through exposure to data and a training process that adjusts weights based on errors, the network develops internal representations: lower layers learn simple features (edges, textures, basic patterns), while higher layers compose these into increasingly abstract concepts (objects, relationships, categories). This is representation learning — the network discovers its own schemas rather than having them programmed in.
The architecture of the training process determines the quality of the learned representations. A network trained on biased data builds biased schemas. A network trained with too little diversity builds brittle schemas that fail on edge cases. A network with insufficient depth builds representations that are too shallow to capture complex patterns. The schemas that emerge are only as good as the construction process that produced them.
This maps directly to human schema creation. Your "training data" is your experience, your reading, your conversations, your environment. Your "architecture" is the mode you default to — inductive, deductive, analogical, social, or experimental. Your "loss function" — the signal that tells you when a schema is wrong — is the feedback you pay attention to. If your feedback loops are weak (Mode 4 social absorption, where the only signal is group agreement), your schemas will be confident but uncalibrated. If your feedback loops are strong (Mode 5 experimental construction, where reality provides direct correction), your schemas will be more accurate but harder to build.
The practical application is that AI tools can augment your schema creation process at its weakest points. If you default to Mode 1 (inductive from personal experience), an AI can surface patterns from data sets vastly larger than your personal sample. If you default to Mode 2 (deductive adoption), an AI can generate counterexamples that stress-test the adopted framework against your specific context. If you default to Mode 3 (analogy), an AI can identify where the analogy breaks down — the structural differences between source and target that your pattern-matching instinct glosses over.
But the AI cannot improve a schema creation process you have not articulated. You have to know what your process is before you can augment it. The externalization is the prerequisite.
Protocol: map your schema creation process
This protocol takes your schema creation from unconscious habit to a documented, improvable process.
Step 1: Collect recent construction events. Identify the last three to five schemas you built — beliefs or mental models that did not exist six months ago. These might be about a new tool, a new colleague's motivations, a market trend, a health practice, anything. Write each one as a single sentence: "I now believe X works like Y."
Step 2: Reconstruct the build sequence. For each schema, trace backward through how it formed. What triggered the need for a new model? Was it a problem, a conversation, a book, a failure? What raw material did you use — personal experience, someone else's framework, analogy to something familiar, group consensus, or deliberate testing? Did the model emerge gradually over weeks or snap into place in a single moment? Did you test it before acting on it, or did you start operating on it immediately?
Step 3: Identify your default mode. Look across all three to five cases. Which of the five modes (inductive, deductive, analogical, social, experimental) shows up most often? That is your default. There is no judgment here — every mode has strengths. But every mode also has characteristic blind spots that you are carrying into every schema you build.
Step 4: Name one structural weakness. Based on your default mode, identify one specific weakness in your schema creation process. If you are primarily inductive, your weakness may be small sample sizes. If primarily deductive, untested assumptions. If primarily social, inherited beliefs you have never personally validated. Write the weakness down.
Step 5: Design one modification. Choose one concrete change you will make the next time you need to form a new mental model. If you are inductive, commit to reading one theoretical framework before forming your conclusion. If you are deductive, commit to running one real-world test before trusting the model. If you are social, commit to finding one person outside your group who holds a different view and understanding why. Make the modification specific enough that you will know whether you did it.
This is not a one-time exercise. It is the beginning of a practice — treating your schema creation process as itself a schema that can be examined, tested, and evolved.
From creation process to quality criteria
You now have something most people never develop: an explicit model of how you build mental models. You know your default mode, your characteristic raw materials, your typical sequence, and at least one structural weakness in the process.
But knowing your process raises the next obvious question: how do you know when the process has produced a good result? What makes a schema high-quality versus low-quality? You can follow a rigorous construction process and still end up with a schema that is inaccurate, too narrow, or too vague to be useful.
In L-0323, you will define schema quality criteria — the specific standards by which you can evaluate any schema you hold. Accuracy, predictive power, simplicity, scope. These criteria become the quality control layer for the factory you just inspected. You have mapped the production line. Next, you build the inspection station.