How to Build a Dark Matter Detector

On semantic dark matter, the knowledge we refuse to trust, and the people who can feel what systems can’t.

Originally published on Medium, March 18, 2026.

It was a Saturday in November. Fall of 2025. Warm but not hot, the way southern New Mexico holds its warmth into the season like a body that hasn’t decided to let go yet. Midday sun. Bright. Last day of early voting and we’d be out of town by Election Day, so this was it.

The girls were out. The house was quiet in that way a house gets when you realize it’s just the two of you and neither of you is in a hurry.

I opened the passenger door for Nayomi. Our choreography. She climbed in, reached for the volume knob, and turned down Nathaniel Rateliff. “And It’s Still Alright” filling the cabin before she dialed it back to make room. Sunglasses. Seatbelts. I put it in drive.

Fifteen minutes to the Doña Ana County building if you catch the lights. Twenty if you don’t. In Las Cruces, running yellows is a local physics. Everyone does it. The posted speed is a suggestion. The actual speed is an agreement between neighbors.

The truck was quiet. I need to tell you why that matters. A year earlier, we’d driven to Phoenix in my old truck, thirteen years of road in that frame, and the highway noise was so loud we couldn’t hear each other. Shoulder to shoulder, facing the same direction, and the noise floor swallowed every word. I got a new truck. Not for the features. For the cabin. The environment changed, and now it could hold the weight of a real conversation.

I’d been writing for months. A manuscript called Break the Chain, about cybersecurity, about physics, about the way digital systems fail when they confuse credentials for identity and snapshots for truth. I couldn’t put the pen down. I’d tell Nayomi about it during our evening conversations, the way you narrate the thing consuming you to the person patient enough to listen. She never read every page. She didn’t need to. She’s a generalist, curious across paradigms, and she absorbed it the way she absorbs everything: through proximity. Through attention. Through the patience of living with someone who can’t stop building.

I knew the architecture was complete. I also knew the soul was missing.

So I asked her. Not what do you think of it. Something more open. What does it need?

She gave me the first thing. I asked small questions to make sure I understood. Then I asked, what else. The second came. Then I pressed, gently, for a third. We were just exiting the freeway near the town of Mesilla, where the Gadsden Purchase was signed, where nations once redrew their borders over a table. I knew we had a few minutes left before the county building. The first two were so good I needed the third. It wasn’t as quick for her. The third came from a different place. Not about the manuscript. About the person writing it.

I knew immediately they were important. Not in retrospect. In the moment. Three things. Three different stretches of road. Each one landing while my hands were on the wheel and the November sun was bright on the windshield.

We arrived. Got out. Held hands. Walked into the county building together to go vote, something we do as ritual. The conversation didn’t end with a period. It ended with a physical act: two people making their voices heard, side by side, carrying three ideas between them that neither one had fully processed yet.

This essay is the processing.

I.

Every knowledge system operates because of factors it doesn’t account for. The physicist John David Krakauer calls this semantic dark matter: the invisible contextual forces that make a model work locally without being explicitly represented within the model itself. Like actual dark matter, it exerts gravitational effect. Like actual dark matter, it doesn’t reflect light. You can’t see it in the equations. But without it, the equations don’t produce the universe you observe.

Darwin’s theory of natural selection is semantic dark matter for thermodynamics. Game theory is semantic dark matter for optimization. The auxiliary theory, the thing operating outside the frame, might actually be the deeper theory. What you thought was the primary model might be the special case.

I build fraud detection systems. Twenty-three years in cybersecurity. I’ve watched organizations deploy models that work beautifully in test environments and fail the moment they touch production. Not because the model was wrong. Because the model didn’t account for the invisible forces acting on the environment: shift changes, seasonal hiring patterns, the mood of an analyst at 3 a.m., the undocumented workaround a team invented six years ago and never told anyone about. The model measured what it was designed to measure. It was blind to everything else. And “everything else” was doing most of the work.

This is not a cybersecurity problem. This is the condition of every system that tries to formalize knowledge.

Consider six rooms. I want to walk through them quickly, the way you’d glance through windows on a passing train. In each room, something is being measured with great sophistication. And in each room, something essential is invisible.

AI safety. The models measure alignment through benchmarks, red-teaming, and reinforcement learning from human feedback. What they don’t measure: the social conditions under which training data was produced. The embedded assumptions about who counts as a “human” in human feedback. The communities whose knowledge was extracted for training and who will never sit at the evaluation table. The semantic dark matter of AI safety is that the field needs the problem to remain hard. Entire ecosystems of research, conferences, careers, and funding are organized around the assumption that alignment is unsolved. A solution that fits on a napkin threatens a complexity that justifies an industry.

Healthcare. The models measure patient outcomes, readmission rates, drug efficacy. What they don’t measure: the patient who doesn’t come back because the bus doesn’t run on weekends. The woman who describes her chest pain differently than the textbook expects and gets sent home. The intergenerational weight of being disbelieved by the system that’s supposed to heal you. A nurse on a night shift can feel when a patient is declining before the vitals change. That feeling has no field in the database.

Criminal justice. The models measure recidivism, risk scores, case disposition rates. What they don’t measure: what reentry actually feels like. The moment you step out of a facility with forty dollars and no address and a phone number for a parole officer who manages three hundred people. The gap between a Second Chance Act on paper and a landlord who runs a background check. Someone who has lived this can tell you exactly where the system breaks. The system does not have a column for their testimony.

Food systems. The models measure calories distributed, supply chain efficiency, agricultural yield per acre. What they don’t measure: the grandmother in Chaparral who drives forty-five minutes to the nearest grocery store with fresh produce. The farmworker whose body feeds a nation and whose family qualifies for SNAP. The difference between food access and food sovereignty. Between being fed and choosing what feeds you. A community organizer who has mapped every corner store in a five-mile radius can tell you where nutrition policy fails. No one is funding that map.

Housing. The models measure vacancy rates, median sale price, units permitted. What they don’t measure: the family paying fifty-three percent of their income in rent who technically isn’t homeless. The way a neighborhood changes when every house on the block becomes a short-term rental and the people who lived there scatter. The teacher who commutes ninety minutes because she can’t afford to live where she works. An acequia commissioner can tell you what happens to a community when the water rights change hands. The housing model has no field for water.

Public education. The models measure test scores, graduation rates, student-to-teacher ratios. What they don’t measure: the kid who tests well in the morning and can’t focus after lunch because lunch was a bag of chips. The teacher buying supplies with her own money for the third year running. The school counselor carrying a caseload of four hundred students and doing triage instead of guidance. A parent who has sat through nine IEP meetings can tell you exactly where the system stops seeing their child. The dashboard doesn’t have a row for that.

Six rooms. Same architecture. In each one: a model that measures what it was designed to measure, an invisible field exerting force on every outcome, and a person who can feel that field because they live inside it.

Aldous Huxley called the brain a “reducing valve.” Not an engine of perception but a filter. Mind at Large, he argued, contains far more than any organism can afford to perceive. The valve narrows reality to a survivable stream. What gets through is what the organism needs in order to function. What gets filtered out is everything else.

Every model is a reducing valve. Every database, every risk score, every policy framework. They don’t fail because they’re poorly built. They fail because they were designed to narrow. The narrowing is the feature. And the cost of the feature is borne by whoever exists in the part of reality that got filtered out.

Carl Jung had a word for the filtered material. He called it the shadow. Not evil, not malice. Just everything a system refuses to look at. “Until you make the unconscious conscious,” he wrote, “it will direct your life and you will call it fate.” A field’s shadow is not its ignorance. It’s its investment in not knowing. The things a discipline could see, if it chose to, but has organized itself around ignoring.

Baruch Spinoza, writing three centuries before any of them, described a worm living in the bloodstream. The worm perceives each particle of blood as a whole universe. It cannot perceive the entire body, the circulatory system, the organism of which it is a small part. It isn’t wrong about what it sees. It is incomplete. And its incompleteness isn’t a failure of effort. It’s a condition of perspective.

We are the worm. Every one of us. In every system we build and inhabit. The question is not how to stop being the worm. It’s how to build instruments that sense what the worm cannot.

II.

Here’s where it gets uncomfortable.

The reducing valves weren’t weather. Someone designed them. Someone chose which variables enter the model and which ones don’t. Someone decided that recidivism could be scored and that the score could be trusted and that the people who objected to the score were “anecdotal.” Someone built the dashboard that shows test scores but not hunger. Someone approved the housing report that counts units but not displacement.

Architecture, not weather. Design, not accident. And the people who designed these systems were not villains. They were doing their best with the tools they had, inside institutions that rewarded measurement and penalized ambiguity. The reducing valve isn’t the fault of any single person. It’s the accumulated consequence of a thousand reasonable decisions, each one narrowing the aperture a little more, until the system is elegant and precise and unable to see the thing that matters most.

This is where Krakauer’s inversion becomes essential. He proposes that the semantic dark matter, the invisible auxiliary theory, might actually be the more fundamental one. The dark matter might be the real theory. What we thought was the theory might be the special case. “One could transform the half-empty glass into a half-full one,” he writes, “by reversing their prioritization.”

Consider what that means. It means the nurse’s intuition isn’t a soft supplement to the clinical model. It’s a more complete theory operating with more variables. It means the returning citizen’s experience of reentry isn’t anecdotal pushback on the risk score. It’s a richer data set that the model was too narrow to ingest. It means the grandmother driving forty-five minutes for produce isn’t a statistical outlier. She’s the primary source. The model that missed her is the outlier.

This is not an argument for abandoning measurement. Measurement is essential. The reducing valve exists because the full signal would overwhelm. But when a field treats its model as reality rather than a sample of reality, when it organizes its conferences and funding around the assumption that the model is the primary theory and the lived experience is supplemental, it has confused the map for the territory. And the people living in the territory pay the cost.

Why do fields resist their own dark matter?

Not because the information is unavailable. In most cases, the people who carry the information are standing right there. The nurse. The farmer. The parent. The person who came home from prison and can describe exactly where the bridge collapsed. They are not hard to find.

The resistance is structural. Miranda Fricker describes two forms of what she calls epistemic injustice. The first is testimonial: your knowledge is discounted because of who you are. The second is hermeneutical: your experience exists but there is no shared concept to name it. You feel something the system doesn’t have a word for. You report it and the system files it as noise.

Both forms operate simultaneously in every one of those six rooms. The patient whose pain description doesn’t match the template. The farmworker whose expertise is physical, not credentialed. The formerly incarcerated person whose map of systemic failure is dismissed as bitterness. Their knowledge is discounted (testimonial injustice) and the categories that would make their knowledge legible don’t exist in the system’s vocabulary (hermeneutical injustice).

The field doesn’t resist its dark matter out of malice. It resists because the dark matter, if taken seriously, would require restructuring the field. New variables in the model. New people at the table. New definitions of expertise. The shadow, as Jung understood it, is not the thing a system cannot see. It’s the thing a system has organized itself around not seeing. Because seeing it would change everything.

And so the cost falls on specific bodies.

Not on the field. Not on the institution. Not on the model. On the people who can feel what the system cannot and who persist inside it anyway.

I know something about this. I grew up inside a system that couldn’t sense what was happening. The adults were present. The infrastructure was in place. But the instrument that would have detected what I was carrying didn’t exist in that environment. I learned early what it feels like to hold knowledge that the system around you has no mechanism to receive.

That’s not a metaphor. That’s the origin of everything I’ve built since.

When you can feel what a system can’t, you become a kind of instrument yourself. Your body registers what the model filters out. You walk into a room and sense the temperature before the thermostat does. You read a policy and feel the gap before anyone publishes a study. This is not mysticism. This is Spinoza’s proprioception: the body’s knowledge of itself in space, its awareness of its own position relative to the forces acting on it.

But here’s the cost: the system wasn’t designed to use your readings. You carry the data and the data has no place to land. You flag the signal and the system processes it as noise. Over time, this is metabolically expensive. It’s the weight of perceiving more than the filter was designed to let through, of knowing something is true and watching it be categorized as anecdote, as emotion, as bias, as insufficiently rigorous. Huxley’s reducing valve exists to protect you. When you see through it, when you perceive what it was designed to block, you become vulnerable in a way that people inside the filter are not. They are protected by what they can’t see. You are exposed by what you can.

The nurse working her fourth double shift who knows the patient in 4B is declining but can’t articulate it in a way the chart will accept. The community organizer who has mapped every food desert in the county and watches the state fund another grocery study. The mother who has sat through a decade of meetings for her child and who knows more about the system’s failure points than any administrator, but whose expertise will never be called expertise because she doesn’t have the credential.

These people are not broken. They are instruments the field has not yet learned to read.

III.

So how do you build a dark matter detector?

Not a metaphor. An actual method. Four steps, and each one requires something the field is not currently structured to provide: slowness.

Name the dark matter. Not “there are things we don’t know.” That’s trivially true and usefully empty. Name the specific invisible forces acting on the specific system. In cybersecurity, I can name them: environmental stability, temporal phase, population pressure, adversarial heat. These are the variables my field’s dominant model, Zero Trust, was designed to ignore. Not because they’re unmeasurable. Because the model was built for a world that doesn’t move. I built a framework called the Data Compass to make them visible. Six fields of measurement where there used to be one. It’s one instrument. Every terrain needs its own.

Find the people who already feel it. They are not hard to find. They are the ones whose knowledge gets categorized as “soft,” “anecdotal,” “not scalable.” They are the nurse, the farmer, the returning citizen, the teacher buying supplies, the grandmother driving forty-five minutes. They’ve been carrying the dark matter in their bodies. The first act of building a detector is to ask them what they feel and believe the answer.

Translate feeling into measurement. This is the hard step. Not because the translation is complex but because it requires sitting with someone else’s way of knowing long enough to find the shared language. A community organizer doesn’t speak in variables and confidence intervals. A clinician doesn’t speak in stories and seasons. The translation is not about converting one language into another. It’s about building a third language that neither party had before. This takes time. This is the slowing down.

Make the measurement load-bearing. A new variable that doesn’t change a decision isn’t a measurement. It’s decoration. The dark matter detector works when the field actually restructures around what it reveals. When the housing model includes water rights. When the recidivism score includes the bus schedule. When the food system model weights the grandmother’s forty-five-minute drive as heavily as the caloric throughput of the supply chain. Load-bearing means the structure depends on it. Anything less is theater.

Who builds these instruments?

People at intersections. Gilles Deleuze and Félix Guattari described the rhizome: a network where any point can connect to any other point, with no hierarchy, no trunk, no central root. It is “not a social construction but a spanning feature of reality,” Krakauer writes, “often too complex to subsume within a single framework.” The people who build dark matter detectors are rhizomatic. They live in multiple systems at once. They speak more than one disciplinary language. They’ve been the signal and the noise. They’ve sat in rooms where their knowledge was primary and rooms where it was invisible, sometimes in the same week.

This is not a romantic portrait. It’s a structural argument. The person who has been both the clinician and the patient, both the policy maker and the person the policy acts upon, both the builder and the one the building displaces, that person carries a parallax that no single-domain expert can replicate. They see the dark matter because they’ve lived on both sides of the reducing valve.

I think about Radia Perlman, who invented the Spanning Tree Protocol. The algorithm that prevents every network on earth from collapsing into an infinite loop. She didn’t get a Nobel Prize. She got a patent. She shipped it. And now every network depends on something so elegant it looks obvious in retrospect. That’s the fate of dark matter detectors. When they work, they disappear into infrastructure. The people who needed them breathe easier. The people who built them rarely get credited with the air.

I don’t have a manifesto. I have a practice.

Start with the room you’re in. Whatever system you operate inside, whatever field you serve, whatever model you rely on to make decisions. Ask yourself what it doesn’t measure. Not abstractly. Specifically. What variable would change your next decision if you could see it?

Then find the person who can already feel it. They’re in the room. They’ve been in the room. They stopped raising their hand because the room stopped listening.

Sit with them. Not for an interview. Not for a study. For a conversation. Shoulder to shoulder. Facing the same direction. Let the exchange take as long as it takes. The instrument you’re building isn’t a dashboard or an algorithm. It’s a relationship between what you can see and what they can feel, and the space where those two forms of knowledge produce something neither one could build alone.

This is slow work. It is the slowest work. And it is the only work that produces instruments capable of sensing what the models miss.

I know the urgency. I feel it every morning. Headlines about systems failing at speeds that outpace every governance structure we’ve built. The window closing. The entrenchment calcifying. I feel the pull to build faster, publish faster, ship the next RFC.

But I had a conversation in a truck, on a Saturday, with a person who has been teaching me for years that the most sophisticated instrument I’ll ever build requires me to stop building long enough to ask what I’m missing. And then to believe the answer.

We pulled into the driveway. I shifted into reverse, backed in. Rateliff was still playing. The November sun was lower now, the light changed.

I gathered the shopping bags from the back. Before I reached the door, I could hear them: Stout, barking because he’s the handsome boy and needs everyone to know we’re home. Zia, whose tail flutters like a butterfly, the kind of involuntary joy the body produces when it doesn’t need to think. And Mila. The old lady. Fifteen, sixteen years with Nayomi. Longer than I have been. Mila has seen everything. She doesn’t verify. She knows.

The house was still quiet. The girls weren’t home yet. It was still just the two of us and three dogs with very accurate clocks who knew it was dinner time.

The Saturday that started shoulder to shoulder in a truck ended shoulder to shoulder in a driveway. The conversation from the morning wasn’t finished. I don’t think it’s supposed to be. I think the best instruments are the ones you keep calibrating, the ones that grow more sensitive as the relationship deepens, the ones that never stop revealing what you couldn’t see before.

Crescit eundo. It grows as it goes.

The state of New Mexico put that on its seal and I think most people drive past it the way you drive past anything written in Latin. But I live here. In the Mesilla Valley, next to an acequia that’s been moving water through this community longer than any of us have been alive. And the longer I stay, the more I understand that the motto isn’t aspiration. It’s description. The practice grows. The instrument refines. The dark matter doesn’t go away, it never will, but your capacity to sense it deepens if you keep showing up, keep asking, keep sitting beside the person who can feel what you can’t.

Someone told me three things on a Saturday in November. I’m still processing them. I expect I will be for a long time.

And it’s still alright.