What I Don’t Know Yet

Every framework is a flashlight. It illuminates what it points at and leaves the rest in the dark.

I’ve spent eleven chapters telling you what Perception-First Design sees. The five layers, the diagnostic, the ethics, the cases, the science. And all of it is real. It works. I’ve watched it work for fifteen years across dozens of projects, from a nightclub in Santa Barbara to Costco pallets to improv theaters to the site you’re reading this on. The pattern holds.

But a framework that doesn’t name its own blind spots isn’t confident. It’s careless. So this final chapter is about what I don’t know, what PFD doesn’t see, and what I haven’t figured out yet. Not as a performative gesture of humility. As an honest accounting from someone who takes this seriously enough to tell you where the edges are.

---

The WEIRD Problem

The evidence base for PFD is predominantly WEIRD: Western, Educated, Industrialized, Rich, Democratic. That’s the term Henrich, Heine, and Norenzayan coined in 2010 to describe the narrow sample that most psychology research draws from. And it applies directly to everything I’ve built.

The underlying cognitive architecture, predictive processing, working memory limits, first-impression formation, that stuff is expected to generalize across humans. Brains are brains. Clark’s prediction machine model doesn’t stop working when you cross an ocean.

But the parameters are culturally shaped. Which visual patterns signal trust. What counts as “too much” cognitive load. How processing fluency maps to credibility. Whether a face in the hero section reads as warmth or as surveillance. These are not universal constants. They’re learned associations, and they’re learned differently in Osaka than in Ohio.

I’ve applied PFD in Western e-commerce and entertainment. That’s my dataset. Whether the five-layer dependency order holds in collectivist markets, in cultures with non-Latin typographic traditions, in markets where trust signals are fundamentally different from what I’ve calibrated against, that was an open question when I first wrote this book.

Recent cross-cultural research has resolved part of it.

Takahiko Masuda and Richard Nisbett showed in 2001 that East Asian observers attend to background and relational context while Western observers focus on foreground and focal objects. Four years later, Hannah Chua, Julie Boland, and Nisbett replicated this with eye-tracking: the same scenes produce different scanning patterns depending on the viewer’s cultural background.

Joshua Goh and his colleagues found cultural differences in face-processing cortex activation in 2010. And Li-Jun Ji, Nisbett, and Yanjie Su had shown earlier, in 2001, that prediction direction itself varies cross-culturally.

Westerners tend to assume linear continuation. East Asians more often anticipate non-linear reversals.

What this evidence resolves: the cognitive architecture is universal, the calibration is cultural.

The five-layer stack holds across populations. What differs is the parameters. L1 calibration (what triggers trust) and L3 calibration (what predictive priors load automatically) have to be validated against target-market users before any design rolls across regions.

Meta-rule: layer requirements are universal; layer calibration is cultural. PFD cannot claim cross-cultural universality without this step.

What the evidence doesn’t resolve: L2 processing-fluency effects across non-Latin typographic traditions, L4 decision-architecture priors across markets with different default authority structures, and whether the dependency-stack ordering itself holds invariant across high-context vs low-context cultures. Those stay open.

Partially resolved is not the same as resolved. But it’s also not the same as unknown. The architecture-vs-calibration distinction is now a working premise, not a guess.

---

The Measurement Gap

The 5-Minute Perception Audit is the most accessible thing in PFD. It’s the part I can teach a business owner in an hour. It’s directionally correct, fast, and useful.

It’s also not validated.

I mean that in the technical sense. There’s no inter-rater reliability data. I haven’t run a study where two PFD practitioners independently audit the same site and I measure whether they converge on the same layer diagnosis. I haven’t compared audit outcomes against controlled user research to see if the heuristic catches what the empirical methods catch. The audit is based on my pattern library, my sensitivity, my years of doing this. It works for me. I believe it works for others who learn it. But I don’t have the numbers to prove it.

The gap between “expert intuition with a framework” and “teachable, repeatable diagnostic” is real. Right now PFD sits closer to the first one. Getting it to the second would mean running the kind of validation studies that turn a practitioner framework into something publishable in a peer-reviewed journal. That’s work I want to do. I haven’t done it yet.

I’m naming this because I think the honesty is more useful than the spin. A lot of design methodologies claim rigor they don’t have. The “proprietary process” that’s really just “the person who built it is good at design.” I don’t want to be that. If PFD’s diagnostic is going to claim it’s teachable and repeatable, I need the evidence. I’m working on it.

---

First Encounters vs. Visit 100

PFD describes first encounters well. That’s what the five-layer stack was built for. Activation points, first impressions, the mental waterfall from cognitive load to decision architecture. All of it is optimized for what happens when someone arrives at your site for the first time.

But user relationships evolve. And I’m not sure how the framework applies to visit 100.

At visit 100, the prediction errors that activated on visit 1 are now expected. The user’s brain has updated its model. Your site is no longer novel. The fluency that felt trustworthy on first encounter has become invisible, the way you stop noticing the hum of your refrigerator. The activation points that broke autopilot on the first visit are now part of the autopilot.

Does the five-layer stack apply the same way to retention? To long-term engagement? To the moment when a loyal user starts to drift? I think it does, but in a different mode. The dependency order probably holds. But the interventions at each layer would need to be different. You’re not designing for first-impression trust anymore. You’re designing against habituation. Against the slow fade of attention that comes from familiarity.

There’s research on this. Berlyne’s work on novelty and arousal from the 1970s. More recent work on hedonic adaptation. But I haven’t integrated it into PFD in any systematic way. The framework is weighted toward first contact, and I haven’t done the work to extend it to the full lifecycle. That’s a gap, and it matters for anyone using PFD on a product with a long user relationship.

---

The Calibration Problem

Here’s the one that keeps me up at night.

The diagnostic starts with Feel. Arrive at the page. Let the emotional response fire before your conscious brain translates it. “This makes me feel X.” That pre-verbal read is the primary instrument of the entire methodology.

But that instrument is calibrated by my own culture, class, neurotype, and aesthetic history. I named this in Chapter 3: the autism gives me analytical social cognition, the ADHD gives me friction sensitivity. I called it a tuning fork. And it is. But every tuning fork has a frequency, and mine is not universal.

I’m a half-Mexican, half-Ukrainian, college-educated, neurodivergent man who grew up in Southern California and lives in San Francisco. My “feel” is calibrated by that. The things that trigger my friction response, the things that register as warm or cold, the things that read as trustworthy or sketchy, those responses are shaped by my life, not by some objective standard.

I named my neurodivergence as an advantage. Heightened sensitivity to friction. And it is. But heightened sensitivity to friction could also mean under-sensitivity to other things. Social warmth cues that are obvious to neurotypical designers. Cultural signifiers outside my experience. Accessibility needs I don’t personally feel. Emotional tones that register differently for people whose nervous systems are wired differently from mine.

The diagnostic instrument has its own biases. And PFD doesn’t yet have a protocol for calibrating them. I’ve told you to “know your instrument” and “compensate for its tuning.” That’s honest advice. But it’s also incomplete. Knowing your own biases is necessary. It’s not sufficient. You need external input, diverse perspectives, real user data, to catch what your instrument misses. And right now, PFD doesn’t formalize that. The Feel step is personal. The correction for the Feel step’s blind spots is... informal. That needs to get better.

---

Accessibility: Optimization vs. Rights

Accessibility sits in an uncomfortable place in PFD.

The framework handles it through Layer 0 (cognitive load reduction) and Layer 2 (processing fluency). This produces good accessible design in practice. Sites built with PFD tend to be more accessible than average because the framework naturally prioritizes clear hierarchy, readable text, manageable information density, and consistent interaction patterns. The curb cut effect I described in Chapter 3 is real, and it runs through the whole methodology.

But disability justice scholars make a different argument, and I think they have a point. They say accessibility is a rights issue, not an optimization target. And PFD’s lens is fundamentally utilitarian. It frames accessibility through “does this reduce cognitive load” and “does this improve processing fluency.” Those are optimization questions.

They produce good outcomes, but the reasoning is instrumental. You optimize because it converts better, because it reaches more users, because it reduces friction across the board.

The disability justice framing says: you make things accessible because people deserve access. Full stop. Not because it improves your conversion rate. Not because it happens to benefit other users too. Because excluding people from designed experiences is a harm, and the obligation to not cause that harm doesn’t depend on whether it helps your bottom line.

I think both frames are true. I think PFD’s utilitarian lens gets you 80% of the way to good accessible design, and the rights-based frame covers territory the optimization lens never reaches. But I also think there are cases where they conflict. Where the utilitarian calculation says “this accessibility investment doesn’t improve outcomes for enough users to justify the cost” and the rights-based calculation says “doesn’t matter, do it anyway.”

PFD currently lives in the first calculation. The framework gives you tools to build accessible design. It doesn’t give you a reason to build accessible design when the tools say it’s not worth it. That’s a gap. And I’m not sure a practitioner framework can close it. It might require something more like a value commitment that sits outside the framework entirely.

---

The AI Question

And then there’s the question I get asked most often by other designers: what happens when AI generates the design?

PFD’s diagnostic starts with a human body. The Feel step, the pre-verbal emotional read, the “neurodivergent tuning fork” that resonates at the slightest vibration. The methodology is built around the idea that the designer’s own perceptual system is the primary instrument. You feel the design before you analyze it. You feel the friction before you name the layer.

Can an AI do that?

I’m genuinely uncertain. Not as a rhetorical gesture. I don’t know. Current AI systems can analyze visual hierarchy, check contrast ratios, evaluate reading level, flag inconsistent spacing. They can run the mechanics of the Perception Audit faster than I can. But the Feel step isn’t mechanics. It’s the pre-verbal emotional response of a nervous system that has spent decades being shaped by culture, aesthetics, social interaction, and the particular way my brain processes sensory input.

If the diagnostic can be automated, PFD scales beyond the individual practitioner. That’s exciting. But what’s lost when the instrument is no longer a human body? The whole methodology grew from a specific kind of embodied sensitivity. The friction I feel in a cluttered interface. The warmth I notice when a hero section gets the first impression right. The unease I can’t name until I’ve sat with it for ten seconds and let the feeling resolve into language.

An AI running the diagnostic would be running a different diagnostic. Maybe a better one in some dimensions. Faster, more consistent, less biased by the designer’s own calibration. But the pre-verbal feel, the part that catches things before they have names, I don’t know if that transfers. And if it doesn’t, then PFD has a ceiling on how far it scales without a practitioner in the loop. It’s a methodology and tool for practitioners to automate work they can vet, not a system that replaces the practitioner entirely.

I think the future is both. AI handles the analysis, the layer-by-layer audit, the pattern matching. The human handles the feel. But I haven’t figured out where the handoff goes, and I’m not pretending I have.

---

Ontic Occlusion

Cory Knobel introduced me to a concept that applies directly to this chapter: ontic occlusion. The idea that any representation of reality blocks other representations from being seen. Every lens that makes something visible makes something else invisible.

PFD makes perception visible. It makes the gap between how things are and how things are experienced into something you can diagnose and design for. But in doing that, it occludes things.

Structural power dynamics. PFD’s ethical test says “align perception with reality.” But it doesn’t interrogate who decides what “reality” is. When Simply Smart Home’s product page undersold a good product, the reality was clear. But what about a company whose product is mediocre and whose “reality” is ambiguous? The designer choosing which reality to align with is itself a power move. PFD doesn’t surface that.

Individual vs. collective effects. PFD models one user’s cognitive journey. It doesn’t model what happens when thousands of users are conducted through the same waterfall simultaneously. Individual perception optimization can produce collective harms: engagement loops, attention extraction at scale, homogenization of choice. The methodology’s unit of analysis is the user. The effects that emerge at population scale are outside its frame.

The fluency trap. Processing fluency is Layer 2’s core mechanism: if it’s easy to process, it feels true. That mechanism is value-neutral. It works for honest brands and dishonest ones. A fluent lie is more persuasive than a disfluent truth. PFD’s ethical tests catch deliberate manipulation. But they don’t address the structural problem: by optimizing for fluency, the framework makes it harder for users to engage the critical evaluation that would catch deception. The very thing that makes PFD effective (reducing prediction error, keeping users on the path) is the same thing that can prevent them from stopping to question.

That’s ontic occlusion baked into the mechanism. And I don’t have a clean answer for it.

---

I started this book with a story about a barback getting chewed out by his boss on a Tuesday night. The boss saw one snapshot, formed an impression, acted on it like it was the full picture. His perception didn’t match reality.

Twelve chapters later, I’ve given you a framework for seeing those gaps and designing for them. Five layers. A diagnostic. An ethical test. Specific cases where it worked. Specific science that explains why.

And now I’ve given you the list of things I haven’t figured out. The cultural limits of my evidence base. The measurement gap in my diagnostic. The temporal dynamics I haven’t modeled. The biases in my own instrument. The accessibility frame that my utilitarian lens can’t fully hold. The AI question that could either extend the methodology or break its foundational assumption. The structural blind spots that the framework occludes by the very act of making perception visible.

A living methodology names its edges. That’s what this chapter is for. Not to undermine what came before, but to mark the places where the next work needs to happen. Mine, or someone else’s.

I believe this framework is true. I believe the layers are real, the dependency order is real, the diagnostic works. I believe the science supports it and the results demonstrate it. I also believe that the version of PFD I’ve described in these twelve chapters is an early version of something that could be much more complete. The bouncer who got promoted because he noticed a perception gap, the designer who turned that instinct into a methodology, the person writing this sentence right now, we’re all the same person at different points in the same long process of figuring out how humans actually perceive, and what to do about it.

Design is intentionality from our minds into reality. That’s stupidly profound when you sit with it.

Design is a lens of empathy to solve with. For whom. Always for whom.

“Life could be worse, Calvin.”
“Life could be a lot better, too!”

Bill Watterson, Calvin and Hobbes

---

Afterword

Thank you for reading this. The whole thing. That means more than I can put into a sentence. If something landed, or if something didn’t, send me an email and tell me your thoughts. I’d love to hear from you.

This book started as notes in a bound notebook and became a methodology and then became whatever this is. It wouldn’t exist without the people who believed in me while I was figuring it out. My loved ones, my chosen family, the friends who listened to me talk about processing fluency at dinner and didn’t change the subject. You know who you are. Thank you for being there along the journey and for believing in me when the evidence was mostly just enthusiasm and a spiral of half-formed ideas.

And to you, the reader: I hope something in these twelve chapters changed how you see. That’s all I was trying to do.

Stefan Kovalik
San Francisco, 2026

---