Your most confident beliefs are the ones most in need of testing
You carry hundreds of schemas — mental models about how the world works. Some you formed through careful study. Most you absorbed through experience, repetition, and social reinforcement. A few you adopted because they were convenient. And nearly all of them share one property: you have never deliberately tested them against reality.
This is not a minor gap. It is the central vulnerability in how humans think.
A schema that has never been tested operates at the same epistemic level as a guess. It may happen to be accurate. It may be wildly wrong. You cannot tell the difference without contact with evidence, because the feeling of certainty is not correlated with actual correctness. Conviction is a psychological state, not an epistemic one. The schema "I work best under pressure" feels just as true whether it is accurate or whether it is a story you tell yourself to justify chronic procrastination. Only a test can distinguish between these.
Karl Popper built an entire philosophy of science around this insight. In Conjectures and Refutations (1963), he argued that the difference between a scientific theory and a pseudoscientific one is not the amount of evidence supporting it — it is whether the theory can, in principle, be proven wrong. A claim that accommodates every possible observation is not knowledge. It is unfalsifiable, and unfalsifiable claims are epistemically empty regardless of how sophisticated or emotionally compelling they sound.
Popper's formulation was aimed at science, but the principle applies to every schema you hold. Your belief about what motivates your team, your model of how your industry will evolve, your assumption about why your last project failed — each of these is a conjecture. The question is whether you treat them as conjectures that need testing, or as settled truths that merely need defending.
The psychology of untested schemas
Jean Piaget's developmental research revealed a mechanism that explains why schemas resist testing. Piaget described two complementary processes by which the mind handles new information: assimilation and accommodation. Assimilation is fitting new experience into existing schemas — interpreting what you encounter through the models you already have. Accommodation is modifying your schemas when new experience cannot be assimilated — when reality forces a structural change in how you understand things.
The critical insight is that the human cognitive system strongly prefers assimilation. Accommodation — genuine schema modification — is triggered only when assimilation fails badly enough to produce what Piaget called disequilibrium: a state of cognitive conflict where your existing model and your experience are in undeniable contradiction. Your mind does not seek out this conflict. It avoids it. Assimilation is comfortable; accommodation is disruptive. Given a choice, your cognitive machinery will stretch, distort, and selectively filter incoming information to preserve existing schemas rather than revise them.
This is why untested schemas persist. They are never forced into contact with evidence that could create disequilibrium. When you never test a belief, you never encounter the contradiction that would trigger accommodation. The schema remains intact not because it is accurate, but because it has been shielded from the kind of experience that would reveal its inaccuracies.
Daniel Kahneman's dual-process model deepens this picture. Your System 1 — the fast, automatic, intuitive processing system — generates and maintains schemas effortlessly. System 1 is a pattern-completion machine: it takes partial information and fills in the rest based on existing models. This is useful for speed but catastrophic for accuracy, because System 1 does not distinguish between a schema that has been validated and one that merely feels familiar. Familiarity and truth feel identical from the inside.
System 2 — the slow, deliberate, analytical system — is what you need for genuine schema testing. But System 2 is lazy. It requires effort to activate, and it tends to accept the outputs of System 1 unless something forces it to engage. Kahneman's research demonstrates that most people, most of the time, do not spontaneously engage System 2 to question the schemas that System 1 produces. The result is a mind full of untested models that feel like knowledge.
The Wason selection task: empirical proof that humans avoid falsification
The most striking experimental demonstration of this tendency comes from Peter Wason's selection task, first published in 1966. Wason presented participants with four cards, each showing a letter on one side and a number on the other. The visible faces show: A, K, 4, 7. Participants were told the rule: "If a card has a vowel on one side, it has an even number on the other." Their task was to identify which cards they needed to flip to test whether the rule was true.
The logically correct answer is to flip A (to check if it has an even number) and 7 (to check if it has a vowel, which would falsify the rule). But across decades of replication, only 10-25% of participants choose correctly. The most common error is flipping A and 4 — seeking confirmation (does the vowel have an even number?) while ignoring the card that could provide falsification (does the odd number hide a vowel?).
This is not a failure of intelligence. Wason's participants included university students, graduate researchers, and professionals. The failure is structural: human cognition defaults to seeking confirmation of existing hypotheses rather than seeking disconfirmation. You look for evidence that your schema is right. You do not look for evidence that it is wrong. And the evidence that it is wrong is precisely the evidence that matters most, because a single genuine counterexample is more informative than a thousand confirmations.
Wason designed the task explicitly to test whether people could reason like Popper — by looking for falsifying evidence. The answer, robustly replicated across over 2,000 studies, is that they cannot do it naturally. It is a skill that must be cultivated deliberately.
What testing a schema actually looks like
Testing a schema is not abstract philosophizing. It is a concrete practice with identifiable steps.
Step 1: State the schema as a falsifiable claim. Most schemas live in your head as vague impressions. "Remote teams are less effective." "My manager does not value my contributions." "Content marketing drives our pipeline." To test a schema, you must first articulate it precisely enough that evidence could contradict it. "Remote teams at our company produce fewer shipped features per sprint than co-located teams" is testable. "Remote work has downsides" is not.
Step 2: Identify the prediction your schema makes. Every useful schema implies predictions about observable reality. If your schema is correct, what should you see? If your schema is wrong, what would you see instead? Write both predictions down before looking at evidence. This is essential because once you see the evidence, your mind will retroactively adjust the prediction to fit — a phenomenon psychologists call hindsight bias. Pre-registering your prediction locks your schema into a testable form.
Step 3: Seek disconfirming evidence first. Because your cognitive default is confirmation-seeking, you must deliberately counteract it. Ask: "What evidence, if I found it, would force me to revise or abandon this schema?" Then go look for that evidence specifically. This is not pessimism. It is epistemic hygiene. If you search for disconfirming evidence and cannot find any, your confidence in the schema is genuinely warranted. If you only search for confirming evidence, your confidence is an artifact of selection bias.
Step 4: Update the schema based on results. This is where Piaget's accommodation comes in, but done deliberately rather than waiting for disequilibrium to force it. If the evidence supports your schema, note the conditions under which it held — schemas have scope. If the evidence contradicts your schema, revise it. Not "defend it harder." Revise it. The willingness to revise is what separates a person doing epistemology from a person doing rationalization.
Why most people skip this
Testing schemas is uncomfortable for several reasons, and naming them reduces their power.
Psychological investment. The longer you have held a schema, the more decisions you have made based on it. Testing it risks invalidating not just the belief but every action that depended on it. The sunk cost is not monetary — it is identity-level. "I have spent fifteen years building teams based on this hiring philosophy" makes the philosophy feel load-bearing for your self-concept, not just your team.
Social cost. Many of your schemas are shared with your peer group, your organization, or your professional community. Testing a schema that your entire team treats as an axiom can feel like a social betrayal. The schema "our industry is relationship-driven" may survive in your organization not because it has been validated but because challenging it would require challenging the people who built their careers on it.
Effort asymmetry. Holding an untested schema requires zero effort. Testing it requires work — defining the claim, identifying evidence, seeking disconfirming data, updating your model. The payoff is better thinking, but the cost is immediate. Humans discount future benefits and overweight immediate costs. This is why "I should test my assumptions" is one of the most widely agreed-upon and least frequently practiced principles in professional life.
None of these reasons make schema-testing less necessary. They explain why it does not happen by default. The practice is valuable precisely because it is not automatic.
The PKM dimension: notes as untested hypotheses
If you maintain a knowledge management system — notes, a Zettelkasten, a digital garden, a Second Brain — every note in it encodes a schema. When Sönke Ahrens describes the Zettelkasten method in How to Take Smart Notes, he emphasizes that notes should be written in your own words as claims about how things work. Each note is, in effect, a hypothesis. But Ahrens also stresses that the value of the system comes from connecting notes and seeing where connections create tension — where two notes contradict each other, where a new note challenges an assumption embedded in an older one.
This is schema validation built into a knowledge system. The connections between notes are not just for retrieval. They are for testing. When you link a new note to an existing one and notice that they conflict, you have created an opportunity for accommodation — for revising your model rather than merely adding to it. A knowledge system that only accumulates notes without testing them against each other is a collection of untested hypotheses that grows larger without growing more accurate.
Tiago Forte's "Building a Second Brain" framework includes a phase he calls "Express" — using your captured knowledge to create output. This is a form of testing, because the act of expressing a schema in a deliverable form (an argument, a decision, a design) exposes it to feedback from reality. The schema that seemed solid in the privacy of your notes may collapse when you try to use it in a real context. Expression is a validity check that pure accumulation cannot provide.
AI and the Third Brain: validation at machine scale
The entire discipline of machine learning is, at its core, a schema-testing operation. A model is trained on data — it forms a schema of patterns in that data. Then it is evaluated on held-out data that it has never seen. The gap between training performance and test performance reveals how well the schema generalizes versus how much it merely memorized the specifics of its training set.
This is exactly the problem with untested personal schemas. A schema formed from your experience may be "overfitting" — capturing the specific patterns of your particular history rather than the general principles of how the world works. You worked at three startups where aggressive sales tactics correlated with growth, so your schema says "aggressive sales drives growth." But that schema was trained on three data points in a specific market era. Its test-set performance — how well it predicts outcomes in new contexts — is unknown because you have never evaluated it outside your training set.
The AI model evaluation framework maps directly onto personal epistemology. Training data is your past experience. Validation data is deliberately sought evidence that you have not seen before. Test data is the real-world outcome when you act on your schema. The discipline that machine learning engineers apply to their models — never trusting training performance alone, always validating on unseen data, always measuring generalization — is the discipline you need for your own mental models.
When you use AI as part of your thinking infrastructure, you can leverage this parallel directly. Ask an AI to steelman the counterargument to your schema. Ask it to find cases where your model's predictions would fail. Ask it to identify the implicit assumptions your schema depends on. This is not outsourcing your thinking. It is using a tool that does not share your confirmation bias to stress-test models that your own psychology would prefer to leave unexamined.
But the AI cannot decide whether a schema is worth testing, which schemas are load-bearing for your decisions, or what counts as a meaningful counterexample in your specific context. Those judgments require your understanding of what the schema is for. The human contributes the stakes. The AI contributes the dispassion. Together, they create a validation loop that neither can sustain alone.
The bridge to falsifiability
This lesson establishes the principle: untested schemas are hypotheses, not knowledge. But stating that a schema must be tested raises an immediate follow-up question: what does it mean for a schema to be genuinely testable?
Not all schemas are testable in their current form. "Everything happens for a reason" is a schema that no possible observation could contradict, because any event can be retrospectively assigned a "reason." It is not wrong — it is unfalsifiable, which in Popper's framework means it is not even in the category of claims that can be wrong or right. It is epistemically inert.
L-0282 takes up this question directly: what makes a schema falsifiable, and what do you do with schemas that are not? The move from "schemas must be tested" to "schemas must be testable" is the move from aspiration to method — from knowing that validation matters to knowing how to structure your beliefs so that validation is possible.
You have spent fourteen phases building cognitive infrastructure: capturing thoughts, organizing them, connecting them, structuring them hierarchically. Phase 15 asks the question that all that structure exists to serve: is your model of reality actually accurate? The answer does not come from conviction, consistency, or elegance. It comes from contact with evidence. Testing is how you get there.
Sources
- Popper, K. (1963). Conjectures and Refutations: The Growth of Scientific Knowledge. Routledge.
- Piaget, J. (1952). The Origins of Intelligence in Children. International Universities Press.
- Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
- Wason, P. C. (1966). Reasoning. In B. M. Foss (Ed.), New Horizons in Psychology. Penguin.
- Ahrens, S. (2017). How to Take Smart Notes. CreateSpace Independent Publishing Platform.
- Forte, T. (2022). Building a Second Brain. Atria Books.
- Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175-220.