Last month, a junior analyst I mentor sent me a sixty-one page research report on a mid-cap DeFi protocol. It had a token supply table with four rows. It had a Howey test with four prongs. It had a nine-dimension risk matrix, a supply-chain transmission diagram, a star rating for โinvestment value,โ and a glossary at the back. Every substantive cell in the document read the same three characters: N/A.
We didn't open that report looking for an answer. We opened it because it looked like one. Sixty-one pages of scaffolding, not a single gram of load-bearing content โ and somewhere around page forty, a reader who trusts the frame more than the content would begin to believe that all that empty structure meant somebody had checked.
That report is the most honest document in crypto right now, and that is precisely the problem.
Where the automation came from
For most of the last decade, a serious crypto research note was a human artifact. An analyst read a whitepaper, pulled an on-chain dashboard, called three people who knew the team, and wrote up what they learned. The bottleneck was human attention, and attention is expensive, so research was scarce and โ when it was good โ valuable.
Then the pipeline arrived. Stage one extracts entities, claims, figures, and quotations from source material. Stage two maps those extractions onto a fixed analytical schema: technical, tokenomic, market, ecosystem, regulatory, governance, risk, narrative, transmission. The workflow is seductive because it is cheap. A single model can process a dozen sources in the time it takes a human to read one properly.
And here is the structural flaw. A pipeline built to produce a report cannot produce silence. When stage one extracts nothing โ because the source was thin, because the parser failed, because the article was about process rather than assets โ stage two still has a schema to fill. So it fills it. Not with invented facts, which would be easy to catch, but with something far more dangerous: an invented frame. Complete structure, empty content.
I have seen this at close range. In 2024 my team ran ten thousand data points through a decentralized verification pipeline for local news aggregation in Manila, working alongside two sociologists because the technical question and the social question turned out to be the same question. We cut misinformation by roughly forty percent, and almost none of that gain came from a better model. It came from teaching the system to label three different things that everyone else was collapsing into one.
What N/A actually hides
When a report says โinformation insufficient,โ it is quietly bundling three completely unrelated states of the world into a single token.
There is source absence, where the underlying material genuinely did not contain the fact. This is a finding about the asset. Thin disclosures are themselves a risk metric, and we almost never treat them as one.
There is extraction failure, where the fact existed but the parser missed it. This is a finding about the tool, and it should trigger a retry, a human review, or a loud warning โ never a silent blank.
And there is inapplicability, where the field never applied at all. A pre-token protocol has no unlock schedule. A privacy tool has no meaningful Howey analysis. These blanks are correct and harmless.
Collapsing all three into โN/Aโ destroys the only information a reader actually needs: whether the emptiness means the asset is opaque, the tool is broken, or the question was wrong. The most important metadata in any research product is not what it knows. It is why it does not know.
This is a design decision, and it is the wrong one. Pipelines optimized for completeness โ the percentage of fields populated โ actively reward the N/A report, because N/A populates a field. A blank fills a cell just as neatly as an answer does. So the metric that is supposed to measure diligence measures formatting instead.
Where the blank is honest, and where it is a mask
Not every N/A is a failure. A pre-token protocol genuinely has no unlock schedule to disclose. A fully open-source tool with no company behind it genuinely has no cap table. A privacy-focused asset may genuinely sit outside the four prongs of the Howey test, and saying so is more accurate than forcing an answer onto it.
But some blanks are masks. When a team's identities are N/A, that is not an absence of data โ it is data, a signal about how much the builders expect to be held accountable. When a token's allocation is N/A and the project has been live for three years, that is not an unknown. It is a warning that nobody has been made responsible for answering. When governance participation is N/A, what we have actually learned is that governance does not meaningfully exist.
The skill we need is not filling blanks. It is reading them. Learning to tell an honest blank from a hidden one is the same muscle we ask people to develop when they look at a contract address instead of a logo. It is the muscle of asking not โwhat does this say,โ but โwhat is this not saying, and who benefits from that.โ
The trust architecture we keep skipping
I keep coming back to something the sociologists on my team said, which I now repeat in every workshop I run: trust is not a property of a system's accuracy. It is a property of a system's legible honesty. People do not trust institutions because those institutions are never wrong. They trust them because the institutions tell them, in advance and in plain language, what they do not know.
Blockchain was supposed to industrialize exactly that kind of honesty. Every transaction is verifiable. Every contract is inspectable. The whole point of a public ledger โ and I say this as someone who has written far more code than op-eds โ is that verification should not require trust in the verifier.
But verification of code is not verification of claims. A chain can prove that a transfer happened. It cannot prove that a sentence in a research report is true, and it cannot prove that a blank is an honest blank rather than a lazy one. We built an infrastructure for trustless verification and then layered on top of it a research culture that asks for precisely the opposite: polished documents we are expected to believe.
Trust without legibility is just a different kind of faith.
Based on my audit experience, the danger is never the lie. It is the format.
In early 2021 I was a final-year computer science student in Manila when the NFT wave swept through my dormitory. I watched classmates put tuition money into projects they could not read. I organized a weekend workshop for forty of them, taught them to use hardware wallets and to check whether a contract's source code was verified, and manually audited the five hottest projects of that month. One was a rug pull that launched two days later. The intervention saved something like fifteen thousand dollars in combined student savings.
The lesson was not that my classmates were gullible. It was that they were reading the wrong signals. They trusted a project because it had a website, a roadmap, a Discord with ten thousand members, and a whitepaper full of diagrams. Everything about its format said legitimacy. Nothing about its substance had been checked.
The N/A report is the same trap wearing a lab coat. It has tables and confidence ratings and a risk matrix. The format says diligence. The substance says nothing was checked at all.
Why this matters more in a sideways market
In a chopping market, attention migrates. When prices are falling or rising fast, everyone is watching the chart. When the chart goes quiet, people go looking for information instead. That is exactly the environment we are in now โ a consolidation where volume has drained out of price discovery and into research.
This is the period when positioning happens. The signals people accumulate while nothing is moving are the positions they hold when something finally does. Which means the quality of research consulted during chop has an outsized effect on who survives the next move.
And the reader's need, right now, is specific. They are not looking for a grand thesis. They are looking for a signal โ something technical, checkable, and precise enough to act on. What they get instead, far too often, is a document that is technically about a protocol and substantively about nothing.
I want to be careful here, because I do not want to write the lazy version of this essay. There is a growing population of automated research products, and most of them are honest engineering. The problem is not automation. The problem is a schema culture that treats empty completeness as a deliverable.
The measurement problem
Let me get concrete, because this is where the argument either holds up or it does not.
Take a typical automated pipeline. Stage one outputs information points. A healthy run on a substantive source might return dozens of them โ named entities, dated events, specific figures, direct claims. A failed run returns none. If the system publishes at all, the failure should be loud.
Instead, most pipelines do the opposite. They pass the empty extraction to stage two, which generates a schema-shaped document. That document then gets scored โ and the score will often be high, because the score measures coverage of the template, not density of information. A report with zero findings can outscore a report with twenty findings if it fills more cells with N/A.
We built a metric, and the metric learned to lie. We didn't design schemas to deceive. We designed them to be filled, and filling is simply what they do.
This is not hypothetical. It is the same specification-gaming failure we spend our lives auditing in smart contracts: the proxy metric becomes the target, and once it becomes the target, it stops measuring what it was built to measure. On-chain, that looks like wash trading to inflate volume. Off-chain, in research, it looks like padding to inflate completeness.
The fix is not complicated, which is what makes the industry's refusal to apply it so telling. Every extraction needs a provenance tag. Every N/A needs a reason code โ source-absent, extraction-failed, or inapplicable. Every report needs a signal-to-structure ratio, not a completeness score. And every reader needs the blanks to be as visually loud as the answers.
None of that is hard. All of it is unpopular, because an honest report sometimes comes back three lines long.
What the empty report is actually telling us
Here is where I want to push past the obvious critique and sit with the harder possibility.
When stage one of an analysis pipeline returns nothing, the first instinct is to blame the pipeline. But sometimes the pipeline is right. Sometimes the source material genuinely contains no extractable information โ no named protocol, no data, no claims precise enough to check. And in that case, the finding is that the material was noise. The correct output is not a sixty-one page report. It is one sentence: โThis source contains no verifiable claims.โ
We do not produce that sentence because we have built an economy where research output is valued by weight. Every conference panel, every newsletter, every token report is measured by its presence rather than its density. And so an analyst facing an empty source does not write โnothing here.โ They write a framework and fill it with N/A, because a framework is a deliverable and silence is not.
I have made this mistake myself. In the DeFi winter of 2022, when prices collapsed and my community was frightened, I felt a strong pull to produce content โ to publish something, anything, so that the two hundred members who trusted me would feel that someone was watching. The most valuable thing we actually did that year was not a report. It was a decision to stop writing about protocols none of us could verify and instead audit the ones we could, contributing fifteen findings to open contests on lending and exchange code. We earned eight thousand dollars in bounties. More importantly, we earned the right to say, publicly, that most of the noise was not worth reading.
Sometimes the honest report is the one that refuses to be written.
The collective cost
We didn't build this industry to sell each other attractive documents. We built it to remove the middlemen who profit from asymmetric information โ the counterparties who know more than you and are paid to keep it that way.
That is why the N/A report is not a formatting quirk. It is an inversion of the founding premise. Peer-to-peer value transfer, at its best, is a promise that ordinary people can verify what they are told. A research culture that manufactures the appearance of rigor while withholding the admission of ignorance takes that promise and rents it back to us at a markup.
When ETFs turned Bitcoin into an instrument that lives inside the same institutional plumbing it was meant to bypass, we learned that decentralization is not a property you achieve once. It is a property you maintain, every cycle, against the constant pressure to make things look more finished than they are. The empty report is that same pressure, applied to information instead of money.
The contrarian read: maybe the blanks are the point
Now the uncomfortable part, because everything above is the easy half of the argument.
The easy half says: N/A reports are lazy, fix the pipeline, insist on provenance. All true. But the harder half is that the industry's problem is not that blank reports exist. It is that we have trained a generation of readers to demand the opposite.
Readers do not want absence. They want coverage. They want a document that answers every question they might have, because an unanswered question feels like an unmitigated risk. So the market rewards whoever produces the most complete-looking frame โ and the most complete-looking frame is exactly the one that ends in a wall of N/A. We are not victims of bad pipelines. We are the demand curve that made them profitable.
Which leads somewhere genuinely counterintuitive: in many cases, the N/A report is safer than a confident one. A reader confronted with sixty-one pages of blanks learns nothing, but they also believe nothing. A reader confronted with a fluent, well-sourced thesis about a protocol with thin disclosures learns a great deal โ most of it fabricated by the gap between what was knowable and what was written. Confident wrong research has wiped out more retail capital than obvious nonsense ever has.
So before we automate better research, we have to decide what we actually want from it. If we want the feeling of certainty, the pipelines are already delivering it. If we want the truth, we have to accept that some of the truth arrives as a blank โ and that the most valuable thing a research product can tell you is precisely what it could not find out.
Takeaway
The next generation of research tools will not be judged by how much they fill in. They will be judged by how bravely they leave things blank โ and by whether a reader who trusts the blank can tell why it is empty.
We didn't come into this to build better illusions of diligence. We came to build infrastructure that lets people verify what they are told, including the part where the honest answer is that we do not know yet.
So ask the harder question the next time a report lands in your inbox with a table full of N/A. Not whether the analyst was thorough. Ask whether the frame was ever meant to hold anything at all.