On a Tuesday morning in late February, I sat down with a cup of coffee that went cold before I noticed, and I read forty-seven governance submissions to a single DAO proposal. Forty-seven. The proposal itself was mundane โ a routine treasury allocation of roughly $2.3 million to fund a developer grant program for the coming two quarters. What mattered was not the proposal. What mattered was that every one of those forty-seven submissions carried the same title: "Comprehensive Analysis: Proposal 42." Every one ran between two and four thousand words. Every one opened with a paragraph about "the evolving landscape of decentralized finance," moved through a section helpfully titled "Key Considerations," and closed with a recommendation that could be summarized in a single sentence โ proceed, but monitor closely.
I finished the last one at four in the afternoon. And I realized, with a kind of cold clarity I have learned to trust over twenty-seven years in this industry, that I had learned absolutely nothing.
Not one of those forty-seven analyses contained a number that was not already printed in the proposal itself. Not one referenced a comparable grant program, a historical disbursement, a developer retention rate, an on-chain signal of any kind. They were, in the precise language of information theory, pure noise dressed in the grammar of diligence. They looked like work. They performed the ritual of work. And they carried, in the technical sense, zero information gain.
This is the essay I have been avoiding for two years. I am writing it now because the problem is no longer a curiosity. It is infrastructure.
What I am describing is not laziness. It is a market. There is now an entire economy โ tools, freelancers, delegate services, even DAOs whose sole product is the production of analysis-shaped text โ built to manufacture the appearance of diligence without the substance of it. And in a sideways market, where nothing is moving and everyone is waiting for a signal, this economy becomes the dominant form of discourse. When there is no trend to ride, the industry fills the silence with the sound of its own thinking. The trouble is that most of that sound is echo.
Let me be careful here, because the temptation in an essay like this is to wave the word "AI" around like a torch and call it an explanation. That would be too easy, and too flattering. The empty analysis I read in February was not primarily a machine's fault. The template came first. The machine simply learned to fill it faster.
To understand how we arrived at a governance forum where forty-seven people can produce three thousand words apiece and transmit nothing, you have to understand the incentives that built the room before anyone automated it.
The first incentive is volume. In the delegate economy that grew up around major protocols โ Arbitrum, Optimism, Uniswap, and the dozens of smaller DAOs that copied their structures โ influence is often measured in activity. A delegate who votes on every proposal and posts a rationale for each vote looks more engaged than a delegate who votes on three proposals a year and posts a paragraph. The metric rewards the appearance of participation, not its value. So the rational delegate, acting perfectly in his own interest, generates text. He generates it because text is the visible currency of governance, and because the platform that hosts the forum does not, as a rule, measure whether any of it was worth reading.
The second incentive is risk. A governance analysis is a liability document as much as it is a research document. If I write "this grant program will fail" and it succeeds, I look foolish. If I write "this grant program will succeed" and it fails, I look complicit. The only safe position is the one that hedges โ proceed, but monitor closely โ which is precisely the position that says nothing. The template is not a failure of intelligence. It is a rational response to a system that punishes commitment and rewards caution dressed as depth.
The third incentive is the one that arrived most recently and most quietly, and it is the one I have spent the last year trying to name. It is the collapse of the cost of producing plausible text. When generating a competent-sounding analysis was expensive โ when it required a person who had read the code, pulled the data, and formed a view โ the volume of analysis was naturally bounded by the scarcity of analysts. That scarcity is gone. The cost of producing the appearance of analysis has fallen to nearly zero, while the cost of producing the substance of analysis has not fallen at all. The gap between the two is the space in which an entire industry of noise now lives.
Here is the technical crux, and I want to state it plainly, because everything that follows depends on it: information is not the same as words. Information is the reduction of uncertainty.
This is not a metaphor. It is a mathematical definition, and it has teeth. If I tell you that a proposal exists, I have reduced your uncertainty by some measurable amount. If I then tell you that the proposal is about a treasury allocation, I have reduced it slightly more. But if I tell you, in a thousand elegant words, that the proposal should be evaluated on its merits and monitored over time, I have reduced your uncertainty by exactly zero. I have transmitted a thousand words and no information. The words are real. The information is not.
When I audit a governance forum, or when I audit my own thinking, I use a test I borrowed from a decade of reading whitepapers that should never have been written. I call it the subtraction test. I take the analysis, and I ask a single question: if I delete this entire document, what do I no longer know? If the answer is "nothing," then the document was not analysis. It was decoration. Apply that test to the forty-seven submissions I read in February and all forty-seven fail. Apply it to most of the research published about most of the tokens that launch in any given quarter and you will find that the great majority of it fails too. We have built a publishing industry whose central product is the sensation of having been informed, delivered without the inconvenience of information.
I want to be fair to the people caught inside this machine, because I have been one of them. In 2017, when I was running the Ethical Ledger workshops in Chicago โ translating smart contract whitepapers into plain language for retail investors who had no business reading them raw โ I made a discovery that shaped everything I have written since. The value of a translation is not that it repeats the original in simpler words. The value of a translation is that it tells you which parts of the original actually matter. A translation that renders every sentence faithfully has done nothing. A translation that tells you "ignore chapters four through nine, they are padding, and the only sentence that changes your life is on page sixty-three" โ that translation has transmitted information. It has reduced your uncertainty about where to look.
That is the work. And that work cannot be automated by a system that has not read the code, because the system does not know which parts matter. It only knows which parts are common. The common parts are the padding. The common parts are precisely the chapters four through nine that any competent translator would tell you to skip.
This is why the emptiness of the template is not an accident. The template is constructed from the most probable sentences in its genre. And the most probable sentences are, by definition, the sentences that tell you the least. To say "the landscape is evolving" is to say the thing that is most likely to be true of any landscape at any time. It is a statement with high probability and low information. A machine optimized to produce plausible text will produce exactly this. A human optimized to appear diligent, in a system that rewards appearance, will produce it too. The two converge, and they converge on the same emptiness, and that convergence is the story.
Now let me turn to the data, because this is where the essay earns its keep, and because I have a professional allergy to describing a crisis without measuring it.
On-chain governance participation has, for the better part of a decade, hovered in a range that any reasonable observer would call a failure. Turnout on major protocol proposals regularly falls below five percent of eligible tokens, and often below two. I have watched this number for six years. It does not move. It is one of the most stable statistics in an industry that prides itself on volatility. And I want to be precise about what it means, because it is routinely misread.
Low turnout is usually described as apathy โ as though the problem were that people do not care. That reading is comfortable and wrong. The problem is not that people do not care. The problem is that the cost of caring has been deliberately set higher than the value of caring. To vote intelligently on a proposal, a token holder must read the proposal, understand the treasury's history, evaluate the proposer's track record, and form a view on a decision whose outcome will almost never be attributable to her individual vote. The expected personal return on that effort is close to zero. So she does not do it. She delegates, or she abstains, or she votes the way a large delegate voted, because that is the cheapest way to hold a defensible position.
The consequence is the thing I have been writing about since I co-designed the governance structure for UnityDAO in 2020, and it is the thing the industry still refuses to say out loud: when turnout is below five percent, the phrase "community decision-making" is a euphemism. The decisions are made by the five percent. And the five percent are, overwhelmingly, the largest holders โ the funds, the venture arms, the founding teams, the professional delegates who hold enough delegated tokens to swing a quorum.
This is not a conspiracy. Nobody has to coordinate. It is a structural outcome, as predictable as gravity. The people who can afford to pay attention are the people who are paid to pay attention, and the people who are paid to pay attention are the people whose attention is bought by the size of their stake. Every governance forum in the industry, including the one where I read forty-seven empty analyses, is a machine for converting capital into attention and attention back into capital. The forty-seven analyses were not written to persuade anyone. They were written to be seen writing. They were the sound of the five percent talking to itself.
And here is the part that should worry anyone who believes in the promise of decentralization. The template and the whale are not two problems. They are one problem wearing two faces. The template is what governance looks like when participation is meaningless โ when the outcome is already determined by stake, the only rational use of a participant's voice is to make it look busy. The whale is what governance looks like when participation is concentrated โ when a handful of holders can move any proposal, the forum becomes a stage on which they perform legitimacy. The empty analysis is the sound the stage makes. It is not a bug in the system. It is the system working exactly as its incentives dictate.
I have spent the last year, through an initiative I call Human-First Protocols, trying to build a countermeasure. The work is unglamorous and it is slow, and I want to describe it honestly, including its limits, because the industry does not need another manifesto. It needs a manual.
The core of the effort is a manual verification layer applied to high-stakes proposals โ the ones where the treasury impact is material and the decision is irreversible. The layer works like this. Every submission to a flagged proposal is scored against a small number of questions that a template cannot answer. What number in this document is not already in the proposal? What precedent does it cite? What would have to be true for the recommendation to be wrong? A document that cannot answer these questions is not deleted and it is not censored. It is labeled. It is marked, visibly and permanently, as containing no information gain. The label is the intervention. We are not trying to silence the noise. We are trying to make the noise visible as noise, so that the reader's scarce attention can be routed to the two or three submissions that actually contain signal.
The early results, across a thousand proposals we reviewed by hand in 2026, were both encouraging and humbling. Encouraging, because once submissions were labeled, the volume of empty analysis in the flagged threads fell sharply โ not because we stopped it, but because being publicly labeled as empty turned out to be a cost that the producers of empty analysis were unwilling to pay. Humbling, because the underlying problem did not move. Turnout stayed low. Concentration stayed high. We had made the noise quieter without making the signal louder, and quiet is not the same as clear.
That is the honest report. Code without compassion is cold, but compassion without code is just a feeling. We built the code. The feeling โ the genuine participation, the sense that a community is deciding its own future โ is still the hard part.
Now I want to argue against myself, because I have learned that the most useful thing an essay can do is attack its own conclusion before the reader does.
There is a strong case that everything I have written here is a complaint dressed as analysis โ that I have spent three thousand words mourning the emptiness of templates while producing, in a different shape, a template of my own. The case goes like this. I have identified a problem that is structural, that is incentivized, and that has persisted for a decade. I have offered a countermeasure whose own results I admit did not move the underlying numbers. Therefore the reasonable reader should conclude that the problem is not solvable by the means I have described, and that my essay is a well-written way of feeling productive about a situation I cannot change. That is a fair attack, and I want to give it the respect it deserves rather than deflect it.
The deflection would be to say that information quality is a long game, that culture changes slowly, that the label is a seed. All of that is true and all of it is the kind of thing a person says when he has no better argument. So let me offer a better one, and it is counter-intuitive.
The most honest output I have ever produced is the phrase "insufficient information."
I mean this literally. In my analytical work, I maintain a discipline that most of the industry abandoned years ago: when the facts do not support a conclusion, I write, in plain language, that the facts do not support a conclusion. I do not fill the gap with scenario analysis. I do not write "on the one hand, on the other hand." I write the four words โ insufficient information โ and I stop. And every single time I have done this, I have been told, gently or not, that it is unhelpful. That a client wants a view. That a reader wants a takeaway. That a report without a conclusion is a report that will not be read.
And I have come to believe that this pressure โ the pressure to always have a conclusion, to always convert an empty input into a filled output โ is the single most corrosive force in this industry. It is what turns a template into a lie. It is what turns a person of integrity into a producer of noise. It is what turns a governance forum into a stage. The willingness to say "I do not know" is the rarest and most valuable asset in a system optimized for the appearance of knowing.
I once spent three weeks on an analysis of a protocol that I will not name, because after three weeks of reading its code and its treasury history, I could not determine whether its reported reserves were real. I had no proof of fraud. I had no proof of soundness. What I had was a gap โ a place where the data simply did not exist to settle the question. I wrote, at the end of the report, exactly one paragraph, and the paragraph said that the reserves could not be verified from public information and that anyone treating them as verified was making an assumption, not an analysis. That was the entire deliverable. Three weeks of work, and the output was a single honest sentence.
I think that sentence was worth more than forty-seven analyses, and I think the industry's inability to produce that sentence is the reason the forty-seven analyses exist. We have built a culture that cannot tolerate an empty output, and so we fill it, and the filling is the corruption. The template is not the disease. The template is the symptom of a deeper refusal โ a refusal to admit that sometimes the correct answer to "what does this mean?" is "there is not enough here to mean anything."
Let me push the contrarian angle one step further, because I do not want to leave the reader with a comfortable conclusion.
It would be tempting to end this essay by blaming AI. It would be popular. It would let every human in the room off the hook. But the empty analysis I read in February was not, in its essence, an AI problem. It was a demand problem. The machine did not invent the template. The machine learned the template from us, and it learned it because the template was already the dominant form of human governance writing. We did not automate analysis. We automated the ritual that had replaced analysis, and we did it because the ritual was all that was ever really rewarded.
If that is true, then the solution is not better detection, better labeling, or better models. The solution is a change in what we reward. And that is a much harder thing to build than a verification layer, because it requires a community to decide that it would rather have one honest sentence than a thousand beautiful ones, and to enforce that preference with its attention and its treasury. No tool can make that decision. Only people can.
Which brings me, finally, to the question I cannot answer, and to the reason I am writing this at all.
The sideways market will not last forever. When it ends โ when the next cycle arrives and capital floods back and the incentives to perform diligence without doing it become enormous โ the industry will face a choice it has faced before and failed. It can let the noise scale with the money, or it can decide, in advance, that the signal is worth protecting even when it is unpopular. I have watched one cycle where the choice was made badly. I have watched another where it was made worse. I am not confident that the third will be different.
But here is what I know. The protocols that survive the next cycle will not be the ones with the loudest forums. They will be the ones where a single honest sentence still carries weight โ where the willingness to write "insufficient information" is treated as a contribution rather than a failure. Those protocols are rare now. They will be the only ones that matter when the noise gets loud.
The question I leave with you is not whether your protocol has good governance. The question is whether your governance can survive an honest silence. When the analysis runs out โ when someone stands up in your forum and says, plainly, that there is not enough information here to decide โ does your system reward that person, or does it reach for the template?
I already know the answer for most of the industry. I am asking whether you want to be part of the rest.