In late January, a research document crossed my desk in Paris. Forty-one pages. Nine analytical dimensions — technical, tokenomic, market, ecosystem, regulatory, governance, risk, narrative, supply chain. Every table rendered in the correct shade of institutional grey. And beneath the baroque facade, the ledger bled nothing, because it contained nothing.
The technical section read insufficient information. The tokenomic section read insufficient information. The risk matrix — that beautiful grid of six categories and eighteen cells — carried the same three words in every row. There was a methodology page. There was a disclaimer. There was a glossary with no terms in it. The conclusion was a request for more input.
I have read bad research for twenty years. I had not, until that morning, read research that was structurally incapable of being wrong. What unsettled me was not the emptiness. It was that nothing in the distribution chain appeared to notice — or, more precisely, that noticing had quietly stopped being anyone's job.
Context
Let me be precise about what changed, because the lazy explanation — that artificial intelligence flooded the market with slop — is the wrong one, and the wrong diagnosis produces the wrong hedge.
Between 2017 and 2021, crypto research had a physical constraint. To write about a protocol you had to read its whitepaper, trace its contracts, and argue with its incentives. The raw material was public, ugly, and unglamorous. In the spring of 2017, I spent four months in my apartment in Le Marais auditing forty-two early Ethereum projects. It was tedious work. It was also the only work that mattered. When I found the recursion flaw in Parity Technologies' multi-sig wallet architecture and sent a risk assessment to three European institutional funds, the value of that document came entirely from the fact that I had opened the code. Nobody paid me for the format. They paid me for having looked.
The 2024 spot Bitcoin ETF approvals changed the demand side of that equation. Institutional allocators now require documentation — for investment committee minutes, for internal audit trails, for regulatory files that must contain something. Compliance departments do not buy accuracy. They buy artifacts. And an artifact that says insufficient information satisfies the file just as well as one that says accumulate, while carrying none of the professional liability.
Meanwhile the marginal cost of producing the artifact collapsed. Templates are free. Language models will generate nine dimensions of plausible-sounding scaffolding in eleven seconds. The cost of producing a document has fallen to near zero; the cost of producing a fact has not moved at all. That asymmetry is the entire story.
Core
Three mechanisms explain why the empty report is now the rational output of the system rather than the accidental one.
The first is format without content. A research framework is a container. When the container can be generated independently of anything inside it, the container becomes the product. I have watched funds pay five figures for a thematic deep dive whose only verifiable claim was a price chart lifted from a public dashboard. The pages were numbered. The fonts were consistent. The bibliography cited three other reports, which cited two threads, which cited a press release. Provenance degrades with each hop, and nobody is measuring the decay.
The second is the asymmetry of accountability. An analyst who publishes a hollow document is never punished — the document was never falsifiable. An analyst who publishes a correct negative call is punished the moment the market disagrees for a quarter. So the equilibrium drifts, slowly and then suddenly, toward documents that cannot be wrong. The risk matrix with eighteen N/A cells is not a failure of the analyst. It is a triumph of the analyst's career management.
The third is the citation loop. Institutional research now cites other institutional research. Each layer adds authority and subtracts information. By the fourth hop, a claim has a logo, a page number, and no ancestor. I have traced these chains by hand. It is astonishing how often the terminal node is a single anonymous post, laundered through three layers of institutional formatting until it looks like consensus.
Here is what makes this maddening rather than merely sad. The chain is always available. On-chain data is the one form of market intelligence that does not require trust in a narrator. Unlock schedules are published. Wallet concentration is queryable. Gas is measurable. Bridge flows are public. When I modeled volatility compression around the 2024 ETF inflows with two colleagues, we did not need anyone's opinion — we needed the ledger, and the ledger answered.
A nine-dimension framework that cannot fill a single cell has not encountered a data scarcity. It has refused to look. The industry has not run out of data; it has run out of the willingness to read it. Volatility is the tax on ignorance, and this particular ignorance is voluntary.
I have seen the same pattern in the two places where it costs the most. The liquidity fragmentation thesis, which has consumed an extraordinary amount of capital and conference time, describes a condition that routing data does not support at the scale claimed — it is a narrative shaped to justify a product, and the product arrived before the problem did. And in the exchange space, intent-based architectures are celebrated for removing MEV from the public mempool; the technical reality is that the extraction has not disappeared, it has been relocated into off-chain solver networks where the same informational advantages operate with less visibility and no public audit trail. In both cases, the format of the argument is impeccable. The substrate says something else.
Pattern recognition is a burden, not a gift. It means you cannot un-see the empty grid once you have counted its cells.
Contrarian
The consensus explanation — AI slop — is comfortable because it locates the failure in a tool. The failure is in the buyer.
Institutions purchased the form of research because form is auditable and content is not. A compliance officer cannot be dismissed for attaching a hollow report to a memo. She can be dismissed for attaching a correct report that loses money. The incentive is not to know. The incentive is to have documented the attempt to know.
Which produces the genuinely uncomfortable inversion: the insufficient information document is the most honest artifact in the stack. It is the only output that accurately describes the process that generated it. The fraud is not the empty grid. The fraud is the full one — the report with every cell populated, every dimension scored, every risk rated, produced by a process that never opened a contract. Liquidity evaporates when trust calcifies, and trust calcifies fastest around documents that look finished.
Takeaway
The next cycle will not reward coverage. There is already too much of it, and it is worth less each quarter. It will reward provenance — the ability to say, precisely, who paid for a claim, what the claim could not have said, and which cell of the matrix was left blank because the author was honest enough to leave it blank.
So the question to ask of any research crossing your desk is not what it concludes. It is what it was structurally forbidden from concluding. The macro does not whisper; it screams in silence, and most of the industry has learned to file the scream as N/A.