Chapter 30: Field Autopsy: Education

"For every complex problem there is an answer that is clear, simple, and wrong." — Attributed to H. L. Mencken

Chapter Overview

In 2004, a systematic review examined the evidence for learning styles — the widely held belief that students learn better when instruction is matched to their preferred learning modality (visual, auditory, kinesthetic, reading/writing). The review, conducted by Frank Coffield and colleagues, examined 71 different learning style models and found that the evidence for matching instruction to learning styles was, at best, weak and inconsistent. Many of the most popular models lacked basic psychometric validity — the instruments used to classify students into learning styles didn't reliably produce the same results when the same student was tested twice.

In 2008, a comprehensive review by Harold Pashler and colleagues, published in Psychological Science in the Public Interest, went further. They applied a rigorous evidential standard: to validate the learning styles hypothesis, you would need studies that (a) classified students by learning style, (b) randomly assigned them to instruction matched or mismatched to their style, and (c) showed that matched students outperformed mismatched students. They found virtually no studies meeting this standard. The few that came close did not support the hypothesis.

In 2020, a survey of higher education faculty in the United States found that approximately 64% still believed that students learn better when taught in their preferred learning style. Among K-12 teachers, estimates of learning styles belief run even higher — 80% to 95% in multiple international surveys.

The idea has been debunked repeatedly, by multiple research teams, using rigorous methodology, published in high-impact journals, over a period spanning decades. It is still taught in teacher training programs. It is still embedded in professional development workshops. It is still one of the most widely held beliefs about learning in the teaching profession.

Learning styles is this book's paradigmatic zombie idea (Chapter 16). It cannot be killed. And the structural reasons for its unkillability reveal everything you need to know about why education is the field most vulnerable to the failure modes documented in this book.

In this chapter, you will learn to: - Identify why education is structurally vulnerable to every failure mode in this book - Analyze learning styles as the paradigmatic zombie idea and understand why debunking doesn't work - Evaluate the evidence gap between research findings and classroom practice - Assess the structural difficulty of education research and its epistemic consequences

🏃 Fast Track: If you're familiar with the learning styles debate, skim section 30.1 and focus on 30.3–30.5, which analyze the structural reasons education is uniquely resistant to evidence-based correction.

🔬 Deep Dive: After this chapter, read Daniel Willingham's When Can You Trust the Experts? (2012) for a practical guide to evaluating education claims, and John Hattie's Visible Learning (2009) for the most ambitious attempt to synthesize education research across hundreds of meta-analyses.

30.1 Learning Styles: The Zombie That Cannot Be Killed

We examined zombie ideas as a category in Chapter 16. Learning styles is the zombie that best illustrates the structural architecture of unkillability — because the reasons for its persistence have nothing to do with the evidence and everything to do with the structure of education as a field.

The Idea

The learning styles hypothesis — sometimes called the meshing hypothesis — claims that students have distinct learning preferences (visual, auditory, kinesthetic, reading/writing — the VARK model is the most popular classification) and that instruction is most effective when it is "matched" to the student's preferred style. A visual learner should receive visual instruction. An auditory learner should receive lectures. A kinesthetic learner should receive hands-on activities.

The idea is deeply intuitive. People do have preferences — some people like reading more than listening, some people like hands-on activities more than abstract discussion. And teachers' daily experience confirms that different students respond differently to different instructional approaches. The meshing hypothesis takes these real observations and makes a specific causal claim: that matching instruction to preference improves learning outcomes.

The Evidence Against

The evidence against the meshing hypothesis is as strong as the evidence against any claim in education:

Coffield et al. (2004): Reviewed 71 learning style models. Found that the most popular models lacked reliability (the same student gets different results on different days) and validity (the styles don't predict learning outcomes). Concluded that the field's use of learning styles was not supported by the evidence.

Pashler et al. (2008): Applied a rigorous evidential standard (the crossover interaction design — matched vs. mismatched instruction with random assignment). Found virtually no studies meeting this standard. The few that did mostly did not support the hypothesis.

Rogowsky et al. (2015): Conducted a well-designed experiment matching instruction to VARK-classified learning styles. Found no benefit of matched instruction. Auditory learners did not learn more from audio presentations; visual learners did not learn more from text.

Husmann and O'Loughlin (2019): Found that students classified as specific learning types did not study in ways consistent with their supposed style — and that studying in one's "preferred" style did not correlate with better outcomes.

Multiple research teams. Multiple methodologies. Multiple countries. Over two decades. The conclusion is consistent: while people do have preferences, matching instruction to those preferences does not improve learning outcomes. What improves learning outcomes is the quality and structure of instruction, regardless of the modality.

Why It Won't Die

The evidence is clear. The belief persists. Why?

Intuitive appeal (the plausible story problem, Chapter 6). Learning styles feels true. People do have preferences. Students do respond differently to different teaching approaches. The idea provides a plausible narrative that explains observable variation — and as Chapter 6 established, plausible stories that explain observed patterns are extraordinarily resistant to evidence, because the experience of the pattern is more vivid than the statistical evidence against the explanation.

Teacher agency. Learning styles gives teachers a concrete, actionable framework: identify each student's style, then differentiate instruction accordingly. This provides a sense of professionalism and expertise — the teacher as diagnostician. Telling teachers that learning styles don't work is not just presenting evidence; it's removing a tool that makes them feel competent and purposeful.

Commercial infrastructure. Learning styles assessments, training programs, professional development workshops, and instructional materials constitute a multimillion-dollar industry. The companies and consultants who sell these products have financial incentives to maintain the belief. This is a smaller-scale version of the capital-sustained error dynamic from Chapter 29 — not at the scale of the dot-com bubble, but operating through the same structural mechanism.

Training perpetuation. Learning styles is taught in teacher preparation programs and professional development workshops. New teachers learn it from experienced mentors. The idea is embedded in the institutional pipeline that produces teachers — which means that debunking it requires changing not just beliefs but training infrastructure.

No cost to believing. This is the crucial structural feature. A teacher who believes in learning styles and differentiates instruction accordingly is not causing harm — they are simply spending effort on an approach that doesn't produce the specific benefit claimed. The instruction may even be better than undifferentiated instruction, not because of style-matching but because differentiation forces the teacher to think carefully about how to present material. There is no visible cost to believing the wrong thing, which removes the crisis trigger (Chapter 19) that might force correction.

🔗 Connection: Compare learning styles to bite mark analysis in criminal justice (Chapter 27). Both are practices without scientific validation that persist in professional use. But they have opposite cost profiles: bite mark analysis produces wrongful convictions (visible, catastrophic cost), while learning styles produces suboptimal instruction (invisible, marginal cost). This is why criminal justice — despite its formidable structural barriers — faces more pressure to reform than education. The cost of being wrong determines the urgency of correction.

🔄 Check Your Understanding (try to answer without scrolling up)

1. What is the "meshing hypothesis" and what does the evidence say about it?
2. Why does learning styles persist despite decades of debunking? Identify at least three structural reasons.

Verify

1. The meshing hypothesis claims that learning improves when instruction is matched to a student's preferred learning style (visual, auditory, kinesthetic, etc.). Multiple rigorous studies have found no support for this claim — matched instruction does not produce better outcomes than unmatched instruction. 2. (Any three of:) Intuitive appeal/plausible story, teacher agency/professional identity, commercial infrastructure sustaining the belief, training perpetuation through teacher preparation programs, and no visible cost to believing (which removes the crisis trigger for correction).

30.2 The Evidence-Practice Gap: What Research Shows vs. What Schools Do

Learning styles is the most dramatic example of education's evidence-practice gap, but it is not the only one. Across multiple areas of educational practice, there is a systematic disconnection between what the research evidence supports and what schools actually do.

Class Size Reduction

The belief: Smaller classes produce better learning outcomes. This is one of the most widely held beliefs in education — among parents, teachers, administrators, and policymakers.

The evidence: The most rigorous study of class size — the Tennessee STAR experiment (Student/Teacher Achievement Ratio), conducted in the 1980s — found that reducing class size from approximately 22-25 students to 13-17 students produced modest improvements in student achievement, primarily in the earliest grades (K-3). The effects were larger for disadvantaged students.

But "modest" is the operative word. The STAR results, while statistically significant, represented small effect sizes. And the cost of achieving those effects was enormous — reducing class size requires hiring more teachers and building more classrooms, making it one of the most expensive educational interventions possible.

Subsequent research has been mixed. Some studies find small positive effects; others find no significant effect. Meta-analyses generally find that class size reduction has a small positive effect, but one that is dwarfed by the effect of teacher quality. Moving from a below-average teacher to an above-average teacher produces dramatically larger gains than reducing class size — and costs far less.

What schools do: Class size reduction has been one of the most expensive education policies of the past three decades. California's Class Size Reduction program, implemented in 1996, cost billions of dollars and required hiring thousands of additional teachers — many of whom were underqualified, because the rapid expansion of the teaching workforce diluted teacher quality. The policy may have produced negative net effects: the small benefit of smaller classes was potentially offset by the large cost of less qualified teachers.

This is the complexity hiding in simplicity problem (Chapter 15). "Smaller classes are better" is a clean, intuitive claim. The reality — "smaller classes produce modest benefits that depend on grade level, student population, teacher quality, and the alternative uses of the same funding" — is complex and unsatisfying. The clean story wins.

Educational Technology

The belief: Technology in classrooms improves learning outcomes. Putting devices in students' hands, providing online resources, and digitizing instruction will transform education.

The evidence: Decades of research on educational technology have produced remarkably ambiguous results. The largest studies — including OECD analyses across multiple countries — have found no consistent relationship between technology investment and learning outcomes. Some studies find positive effects for specific technologies in specific contexts; others find neutral or negative effects. The consistent finding is that how technology is used matters far more than whether it is used — and that many implementations of classroom technology produce no measurable benefit.

What schools do: Billions of dollars have been spent on educational technology worldwide. The U.S. alone spent an estimated $26 billion on EdTech in 2020. School districts purchase devices, software licenses, and digital platforms with limited evidence that these investments improve learning — and limited mechanisms for evaluating whether they do.

This is capital-sustained error (Chapter 29) applied to education. EdTech companies have powerful financial incentives to sell products. School administrators have political incentives to appear innovative. Parents expect technology in classrooms. The narrative — technology = progress = better education — has narrative-market fit regardless of the evidence.

Homework

The belief: Homework improves learning, and more homework produces more learning.

The evidence: The relationship between homework and achievement is more nuanced than most people assume. Research suggests that homework has a positive relationship with achievement in high school, a weaker relationship in middle school, and little or no relationship in elementary school. The type of homework matters — practice and reinforcement homework shows stronger effects than busywork. Excessive homework may produce diminishing or negative returns.

What schools do: Homework practices vary enormously and are largely determined by individual teacher decisions rather than evidence-based policies. Some elementary schools assign substantial homework despite weak evidence of benefit. Some high schools assign homework loads that research suggests are counterproductive. There is no systematic mechanism for aligning homework practices with evidence.

🧩 Productive Struggle

Before reading the next section, consider: Why is education so uniquely vulnerable to the evidence-practice gap? Every field has some gap between research and practice — medicine's famous "17-year bench-to-bedside gap" is well-documented. But education's gap seems wider and more persistent than most. What is it about the structure of education that makes evidence-based practice so difficult to achieve?

Spend 3–5 minutes, then read on.

30.3 Why Education Research Is Structurally Harder

Education's evidence-practice gap is not primarily caused by ignorance, laziness, or anti-intellectualism among educators. It is caused by structural features of education that make rigorous research extraordinarily difficult — and that make the research-to-practice pipeline leakier than in any other field.

The Difficulty of Randomization

The gold standard of causal evidence — the randomized controlled trial (RCT) — is much harder to implement in education than in medicine or psychology.

In medicine, you can randomly assign patients to treatment or control groups, and ethical review boards have established frameworks for when this is acceptable. In education, randomly assigning students to "good teaching" vs. "bad teaching" raises immediate ethical concerns. Randomly assigning students to different schools, teachers, or curricula disrupts families, communities, and the social fabric of schooling. Parents (understandably) resist having their children used as experimental subjects.

The result: most education research relies on observational studies, quasi-experimental designs, or small-scale experiments that may not generalize. The evidence base is structurally weaker than in fields where randomization is standard — which means that claims about educational effectiveness are built on shakier foundations.

The Measurement Problem

What does it mean for an educational intervention to "work"? The answer is not as straightforward as it seems.

In medicine, outcome measures are often clear: survival rates, symptom reduction, biological markers. In education, the outcomes that matter most — deep understanding, critical thinking, creativity, love of learning, long-term knowledge retention, life outcomes — are difficult to measure. What is easily measured — standardized test scores — is a narrow and potentially misleading proxy for genuine learning.

This is the streetlight effect (Chapter 4) applied to education research. Studies measure what is measurable (test scores), and the results are treated as evidence about what matters (learning). But test scores can be improved by teaching to the test, by narrowing the curriculum, or by gaming the assessment system — none of which represent genuine learning improvement.

The Implementation Problem

Even when an educational intervention has been shown to work in a research setting, implementing it in thousands of classrooms across diverse contexts is an entirely different challenge.

Medical treatments, once validated, can be replicated with high implementation fidelity — the drug is the same drug, the dosage is the same dosage, the protocol is the same protocol. Educational interventions depend on teachers — human beings with different skills, beliefs, motivations, and contexts. An instructional approach that works brilliantly with a skilled, motivated teacher in a well-resourced school may fail completely with a different teacher in a different context.

This makes education research findings structurally less generalizable than medical research findings. The intervention is not separable from the person delivering it — which means that "evidence-based practice" in education is a more uncertain proposition than "evidence-based medicine."

The Time Horizon Problem

Educational outcomes play out over years and decades. The effects of a first-grade reading intervention may not be fully visible until high school — or college — or career. Most education research measures short-term outcomes (end-of-year test scores) because measuring long-term outcomes requires the kind of longitudinal tracking that is expensive, difficult, and rarely funded.

This creates a structural bias toward interventions with visible short-term effects (test score gains) and against interventions with important long-term effects (developing curiosity, building resilience, fostering deep understanding). The field optimizes for what it can measure in the short term, which may not be what matters in the long term.

Opinion Density

Education has a structural problem that no other field in this book shares to the same degree: opinion density. Every adult has been a student. Most adults have strong opinions about education based on their personal experience. This makes education uniquely vulnerable to the authority cascade operating through everyone — not just credentialed experts, but parents, politicians, journalists, and the general public.

In medicine, the average person defers to doctors on questions of treatment efficacy. In education, the average person considers themselves qualified to pronounce on what works and what doesn't — because they went to school. This opinion density dilutes the authority of actual research and makes it easier for plausible stories (Chapter 6) to compete with rigorous evidence.

🔄 Check Your Understanding (try to answer without scrolling up)

1. Why is randomization harder in education research than in medical research?
2. What is "opinion density" and why does it make education uniquely vulnerable to failure modes?

Verify

1. Ethical concerns about assigning students to potentially inferior conditions, logistical challenges of disrupting school assignments, and parental resistance to children being used as experimental subjects all make randomized controlled trials harder to implement in education. 2. Opinion density is the structural feature that everyone has personal experience with education and therefore considers themselves qualified to opine on what works. This dilutes the authority of actual research and makes it easier for plausible stories to compete with rigorous evidence — unlike medicine, where the public generally defers to professional expertise.

30.4 The Failure Mode Stack in Education

Education is uniquely vulnerable because every failure mode in this book operates simultaneously — and the structural difficulty of education research means that the correction mechanisms are weaker than in any other field.

Plausible Story Problem (Chapter 6)

Education is saturated with compelling narratives that substitute for evidence: "children are natural learners" (sometimes, under some conditions), "technology transforms education" (no consistent evidence), "small classes are better" (small effect, enormous cost). These stories are intuitive, align with personal experience, and are promoted by interested parties. They persist because the evidence against them is complex and conditional — exactly the kind of evidence that loses to a clean story.

Zombie Ideas (Chapter 16)

Learning styles is the most prominent, but education has many zombie ideas: learning pyramids (the claim that people retain 10% of what they read, 20% of what they hear, etc. — entirely fabricated, no original source), left-brain/right-brain learning (neuroscience does not support this distinction for education), multiple intelligences (Gardner's theory is widely used in education despite lacking empirical validation for instructional differentiation), and the Mozart effect (listening to Mozart does not improve intelligence).

These zombies persist because they offer teachers simple, actionable frameworks — and because the research debunking them doesn't provide equally simple, actionable alternatives.

Replication Problem (Chapter 10)

Education research suffers from many of the same replication problems as psychology — small sample sizes, publication bias toward positive results, researcher degrees of freedom, and insufficient attention to effect sizes. But education research has an additional replication challenge: because implementation depends on teachers and contexts, replicating a finding in a different setting with different teachers is not the same as replicating a chemistry experiment in a different lab.

Incentive Structures (Chapter 11)

The incentive structure of education creates systematic biases: - Teachers are rewarded for student performance on standardized tests, creating incentives to teach to the test rather than to develop deep understanding - Administrators are rewarded for adopting visible innovations (new technology, new programs), regardless of evidence - EdTech companies are rewarded for sales, not for learning outcomes - Education researchers face the same publish-or-perish incentives as other academics, with the additional challenge that education journals have historically been less rigorous about methodology - Politicians are rewarded for decisive action on education, which means adopting popular policies regardless of evidence

Complexity Hiding in Simplicity (Chapter 15)

Education involves extraordinary complexity — every student brings different prior knowledge, motivation, cognitive development, social context, home environment, and emotional state. Good teaching requires navigating this complexity in real time. But the demand for simple, scalable solutions (from policymakers, administrators, and parents) creates pressure to reduce this complexity to slogans: "smaller classes," "more technology," "higher standards," "school choice."

🔗 Connection: Education's failure mode vulnerability is the opposite of the military's (Chapter 28). The military has massive learning infrastructure but structural forces that override lessons. Education has weak learning infrastructure (no systematic mechanism for translating research into practice) AND structural forces that sustain wrong ideas. The military at least learns during crises and then forgets; education often doesn't learn in the first place, because the research is hard to do, hard to interpret, and hard to implement.

30.5 What Would Fix It (And Why It's So Hard)

The structural problems identified in this chapter suggest that "fixing" education's epistemic vulnerabilities requires systemic change, not just better research or better dissemination.

What Would Help

1. Institutional investment in large-scale RCTs. Education needs the equivalent of medicine's clinical trial infrastructure — organizations that design, fund, execute, and publish large-scale randomized experiments on educational interventions. The Education Endowment Foundation in the UK and the Institute of Education Sciences in the U.S. are steps in this direction, but they remain small relative to the scale of the problem.

2. Closing the research-to-practice pipeline. Medicine has clinical guidelines, Cochrane reviews, and continuing medical education requirements that create structured pathways from research to practice. Education has nothing equivalent at scale. Creating systematic mechanisms for translating research findings into teacher training and professional development would address the evidence-practice gap.

3. De-zombification. Actively removing debunked ideas from teacher preparation programs, professional development curricula, and educational materials. This requires institutional will and faces resistance from the commercial interests that profit from zombie ideas.

4. Outcome measurement reform. Developing better measures of educational effectiveness that go beyond standardized test scores — measures that capture deep understanding, critical thinking, and long-term outcomes.

Why It's So Hard

Every proposed fix faces structural resistance:

• Large-scale RCTs are expensive, ethically complex, and politically unpopular (parents resist randomization)
• Closing the research-to-practice pipeline requires institutional infrastructure that doesn't exist and funding that isn't available
• De-zombification threatens the commercial interests of learning-styles consultants, EdTech companies, and professional development providers
• Better outcome measures are technically difficult, expensive to implement, and threaten the existing accountability systems built around test scores

The result is a field in which the diagnosis is relatively clear — education is vulnerable to every failure mode in this book — and the treatment is structurally blocked by the same forces that created the vulnerability.

30.6 Applying the Correction Speed Model

Prediction: Very slow correction. Education's profile is comparable to nutrition science (Chapter 26) — ambiguous evidence, diffuse costs, no crisis trigger, and weak correction mechanisms. The field may be the slowest to self-correct of any examined in Part IV, because it has the weakest evidence base, the weakest research-to-practice pipeline, and no mechanism for crisis-driven correction.

Comparing Education to Other Fields

Education occupies the worst position on evidence quality and correction mechanisms, with the additional burden of the highest opinion density of any field. It is the field that most needs evidence-based practice and is least structurally equipped to achieve it.

📐 Project Checkpoint

Epistemic Audit — Chapter 30 Addition: The Structural Research Difficulty Assessment

30A. Research Difficulty Assessment. How difficult is rigorous research in your field? Rate the challenges: randomization difficulty, measurement validity, implementation fidelity, time horizon mismatch, and opinion density. Does the structural difficulty of research leave your field vulnerable to plausible stories filling the evidence vacuum?

30B. Zombie Idea Inventory. Does your field have zombie ideas — beliefs that have been debunked by research but persist in practice? List them. For each, identify the structural reasons for persistence (intuitive appeal, commercial infrastructure, training perpetuation, low cost of believing).

30C. Evidence-Practice Gap Assessment. In your field, what is the gap between what research supports and what practitioners do? What structural features of your field explain the gap? What would it take to close it?

30.7 Chapter Summary

Key Concepts

• Learning styles as paradigmatic zombie idea: Debunked repeatedly by rigorous research, still taught in 64-95% of educational settings. Persists because of intuitive appeal, teacher agency, commercial infrastructure, training perpetuation, and no visible cost to believing
• The evidence-practice gap: Systematic disconnection between what education research shows and what schools do — class size reduction (modest effect, enormous cost), educational technology (ambiguous evidence, billions spent), homework (nuanced evidence, unsystematic practice)
• Structural research difficulty: Education research is harder than research in most other fields — randomization is ethically complex, outcomes are hard to measure, implementation depends on teachers, and time horizons are long
• Opinion density: Education is the only field where every adult considers themselves qualified to opine based on personal experience, diluting the authority of actual research
• The complete failure mode stack: Every failure mode in this book operates simultaneously in education — plausible stories, zombie ideas, replication problems, perverse incentives, and complexity reduced to slogans

Key Arguments

• Education's vulnerability is structural, not cultural — it is not caused by teacher ignorance or administrative incompetence but by features of the field that make rigorous research, evidence dissemination, and evidence-based practice extraordinarily difficult
• Learning styles persists not because teachers are credulous but because the structural incentives (intuitive appeal, commercial infrastructure, no visible cost) overwhelm the structural correctives (research that is complex, conditional, and hard to translate into practice)
• Education is the slowest field to self-correct because it has the weakest evidence base, the weakest research-to-practice pipeline, and no mechanism for crisis-driven correction
• The irony: the field that studies learning has learned less about itself than almost any other field studied in this book

Spaced Review

Revisiting earlier material to strengthen retention.

1. (From Chapter 10 — The Replication Problem) Education research suffers from many of the same replication problems as psychology — small samples, publication bias, researcher degrees of freedom. But education has an additional challenge: implementation depends on teachers and contexts, making exact replication impossible. How does this affect the field's ability to build a cumulative evidence base? Compare to psychology's replication crisis.

2. (From Chapter 6 — The Plausible Story Problem) Learning styles is a plausible story that substitutes for evidence. Identify the specific features that make it compelling: it explains observable variation (students differ), it provides an actionable framework (differentiate instruction), and it aligns with personal experience (I prefer visual learning). Why are these features so resistant to statistical evidence against the meshing hypothesis?

3. (From Chapter 15 — Complexity Hiding in Simplicity) "Smaller classes are better" reduces a complex, conditional finding (modest effects, primarily in early grades, dependent on teacher quality, at enormous cost) to a clean slogan. Identify two other educational slogans that hide complexity. For each, describe what the research actually shows.

4. (From Chapter 16 — The Zombie Idea) The chapter identifies several zombie ideas in education beyond learning styles: learning pyramids, left-brain/right-brain, multiple intelligences, the Mozart effect. Apply the zombie resilience taxonomy from Chapter 16: which structural features of zombiehood does each one possess?

What's Next

Chapter 30 concludes Part IV: Field Autopsies. Across eight chapters, we have examined how the failure modes of Parts I-III operate in specific fields — medicine, economics, psychology, nutrition science, criminal justice, military strategy, technology, and education. The patterns repeat: authority cascades, sunk cost, zombie ideas, perverse incentives, and structural barriers to correction appear in every field, with variations driven by each field's unique structure.

In Part V: The Toolkit, we move from diagnosis to prescription. Chapter 31: Red Flags provides 15 diagnostic questions you can apply to any claim in any field — the early warning signs that a consensus may be wrong. The remaining chapters build the practical tools for seeing failure modes in action and doing something about them.

Before moving on, complete the exercises and quiz to solidify your understanding.

Chapter 30 Exercises → exercises.md

Chapter 30 Quiz → quiz.md

Case Study: Learning Styles — Anatomy of an Unkillable Zombie → case-study-01.md

Case Study: The Billion-Dollar Gamble — EdTech Without Evidence → case-study-02.md