Populist Realism: A Manifesto

Peasants can be wiser than kings.

The population should rule. Technocrats should advise.

Technocrats know because they read stuff.

Populist Realists know because they do.

Most hard problems sit between those two. You need people close enough to the ground to see the actual details and people far enough back to see the system that produced those details. Pure technocracy loses the ground and eventually the consent of the people who have to live with the results. Pure populism can lose the mechanisms that make large systems work. Populist Realism holds that the population, grounded in the friction of ordinary life, holds the right to rule. Technocrats supply advice, mechanisms, and long-horizon calculations. They do not hold the right to override the population when the advice fails contact with reality. Systems that invert this relationship become unstable.

Core claim

Expertise is real. Credentials are not the same thing as expertise. Distance from consequences is not the same thing as perspective. Technocrats often live in a cleaner world than the one the rest of us inhabit. Their models begin with assumptions that make the math tractable: frictionless spheres, efficient markets, agents who optimize cleanly, data sets that already contain the relevant variables. Those assumptions are useful for some problems. They are false for most of the ones that matter. Realists work with friction, dirt under the fingernails, and probability distributions that have long tails. They prefer rules of thumb that survive contact with those conditions to theories that only hold when the inconvenient parts are assumed away.

Peasants can be wiser than kings because they cannot opt out of the results. Kings and their advisors can rewrite the definition of success so the model still looks correct. The people who must make payroll, raise the birds, answer the radio, or absorb the yield loss cannot. That is why the population rules and the technocrats advise. The useful test is not which side has more degrees. The useful test is which claims keep matching the ground-level results after the people making the claims have had time to be wrong. Reality is the tie-breaker. The people who cannot opt out of the results are the ones who get to declare whether the results match.

What the two teams actually do

Technocrats prefer the Lawyer-level claim. Precise definitions, peer-reviewed papers, institutional consensus, models that assume the world behaves like the dataset. They are strongest on mechanisms that are slow, technical, or hard to observe directly. They are weakest when the dataset is curated by people who share their incentives, when the people collecting the data cannot be punished for being wrong, or when the lived results diverge from the predicted ones for long enough that the divergence itself becomes the signal. Their default is the frictionless case. When reality supplies friction, they often treat the friction as noise rather than as the dominant feature. Their proper role is advice.

Populist Realists prefer the Irish-level claim. Directional accuracy under real constraints, repeated observation of what people actually do, and feedback from people who cannot opt out of the result. They are strongest on detection of drift between stated goals and delivered outcomes. They are weakest when they treat every institutional failure as deliberate conspiracy rather than ordinary incentive failure, or when they ignore mechanisms that really are technical and slow. Because they live with long tails and irreversible costs, they reach for heuristics that limit downside rather than for models that maximize an idealized objective function. Their proper role is rule.

Each team gets one vote on questions of mechanism and diagnosis so that credentials and institutional volume cannot simply outnumber lived observation. The population holds the deciding vote on whether the system is working and on the direction it should take. Technocrats design and calculate inside the constraints the population sets. They do not get to redefine success after the fact so that the model still looks correct. Cross-team agreement is the strongest signal that a mechanism is sound. Persistent disagreement on outcomes is evidence that the advice has detached from the ground.

Ground-level gaps that keep repeating

The difference is not abstract. It shows up in ordinary production systems that most credentialed commentary never has to run.

Take chickens. A broiler and a fryer are not interchangeable animals with slightly different marketing labels. Modern commercial broilers are the product of decades of selective breeding for rapid weight gain, feed conversion, and uniform carcass size under tightly controlled housing and feed regimens. They reach market weight in weeks. Their skeletal and cardiovascular systems are optimized for that schedule, not for long life or free range. A backyard “fryer” or heritage bird is a different genetic package with different feed requirements, disease pressures, growth curves, and processing logistics. Treating them as the same thing produces confident policy recommendations that collapse the moment someone has to actually raise, feed, house, and process the birds at scale.

Or take glyphosate. An overnight ban would not be a simple substitution of one input for another. Large parts of U.S. row-crop and specialty production have organized around it: no-till and reduced-till systems that control erosion and fuel use, timing of applications that match specific weed pressure windows, and yield assumptions baked into equipment, seed choices, and financing. Removing it without a multi-year transition forces more tillage, more passes, different herbicides with their own resistance and labor profiles, and, in many regions, lower yields or higher costs that existing farm balance sheets cannot absorb. The people who write the environmental risk papers rarely carry the operating loans. The people who do already know the cash-flow and equipment constraints.

Or take policing. What many technocratic accounts describe is a relatively clean system of proactive stops, use-of-force events, and demographic disparities that can be adjusted through training, policy language, and oversight. What officers actually spend most of their time on is the residual of every other failed system: domestic calls, mental-health crises, property disputes, traffic crashes, welfare checks, and the paperwork required to document all of it. Clearance rates for many serious crimes remain low because the volume of calls and the constraints on investigation time do not match the model. The street-level sequence of events on a given shift is driven by 911 dispatch and whatever the caller reports, not by a strategic plan written in a journal. The people writing the disparity studies and the people answering the radio are often describing different jobs.

These are normal. Production agriculture, chemical-dependent cropping systems, and municipal policing are domains where the people who must deliver the output every day see constraints that the models treat as adjustable parameters. The models are not always wrong. They are frequently incomplete in a direction that only becomes obvious once someone has to make the system work tomorrow morning. That is why the population that lives with the output holds the deciding vote on whether the advice has succeeded.

Operating rules

Outcomes over credentials. A model that keeps failing its own predictions loses status regardless of who wrote it. A model that keeps matching outcomes gains status even if the person who noticed it has no institutional title. The people who live with the outcomes get the final call on whether the match is real.

Incentives are part of the evidence. Who pays, who is protected from failure, and who captures the upside are facts. Government rarely has the operating expertise it needs, so it reaches out to industry. The industry advisors do not work for the government. That is ordinary regulatory capture, not an exotic exception. Treat the resulting rules as hypotheses that require independent verification, not as neutral expertise.

Close enough for the details, far enough for the picture. Local knowledge without systemic view produces scapegoats. Systemic view without local knowledge produces elegant disasters.

Demand closed-loop systems. An open-loop system has no feedback from results back to the controllers. In control theory that configuration is unstable. Policy, regulation, and institutional design that cannot observe and correct their own errors will drift. Build the measurement and the correction path in from the start, or accept that the system will eventually run away. The feedback must come from the operators and the population who cannot exit, not only from the analysts who can.

Steelman first. State the strongest version of the opposing claim before you dismantle it. If you cannot, you do not understand it yet.

Institutions are tools, not authorities. They accumulate knowledge and they accumulate capture. Both are observable. Longevity is not the same as accuracy. The population retains the right to discard or redesign the tool when the results no longer match the ground.

The necessity of popular rule with technocratic advice

Technocratic models without realist feedback become brittle. They optimize for the frictionless case and then fail when friction appears. Realist heuristics without technocratic mechanisms become myopic. They correctly notice that something is broken and then lack the tools to redesign the underlying process. Neither failure mode is stable. Open-loop institutions that cannot see their own results, or that only see results filtered through the people who benefit from the current rules, amplify both problems.

The correct relationship is hierarchical in one direction only. The population rules because peasants can be wiser than kings: they see the dirt, the long tails, and the irreversible costs that the frictionless models omit. Technocrats advise because some mechanisms really are technical and some horizons really are long. When the advice improves the model but worsens the cash flow, the carcass yield, or the call clearance, the advice loses. When a popular judgment correctly flags a failure but cannot scale, the technocrats are required to design something that does. Each side still supplies one vote on diagnosis so that neither can simply outnumber or out-credential the other. The population decides the direction and declares success or failure because reality is not optional and the frictionless sphere does not exist.

Practical takeaway

Keep the technocrats for the hard technical mechanisms and the long-horizon models. Keep the population for the detection of when those models have detached from what people are actually experiencing, and for the final judgment on whether the detachment has been corrected. Demand that both sides show their work against the same ground-level results: the growth curves of the birds, the cash-flow effects of the input change, the call volume and clearance data of the department. Insist on closed-loop design so that those results can change the next round of decisions. When the literature and the operating reality point in different directions for long enough, treat the literature as a hypothesis to be tested, not a consensus to be deferred to. The people who cannot opt out hold the deciding vote on the test.

The goal is not to replace one priesthood with another. The goal is to make it more expensive to be directionally wrong for a long time and cheaper to notice when the direction has changed. That requires both the people who read the papers and the people who have to make the system work tomorrow. The second group rules. The first group advises. Systems that invert the relationship are open-loop. They do not stay stable.