A due-diligence dossier landed on my desk last Tuesday. The subject: a modular data-availability protocol that closed a $100 million round eleven days earlier. The document had forty-seven fields โ supply schedule, team verification, audit status, Howey test elements, unlock cliffs, sequencer decentralization. Every field carried the same payload: "N/A โ insufficient information." The headers were correct. The tables rendered. The formatting was immaculate. The document contained precisely zero facts.
I have audited smart contracts since 2017, when my team tore through five thousand lines of Rust on a token-issuance module and found a reentrancy path that delayed a mainnet launch by two weeks. I know what an empty finding looks like. It looks exactly like a filled one, which is the entire problem. The audit reveals what the hype conceals โ but only if someone actually ran the audit.
The dossier in front of me was not produced by a negligent analyst. It was produced by a pipeline โ a two-stage machine that decomposes a source into structured fields, then runs a nine-dimension analysis over those fields. The machine worked perfectly. It simply had nothing to chew on. The first stage returned an empty template. The second stage, bound by a rule that forbids inventing facts, dutifully stamped "insufficient information" across every row and shipped it upstream.
That is the honest failure. The dishonest failure is the one nobody sees.
Context: the industrialization of crypto research
Three years ago, "research" in this industry meant a human being reading a whitepaper, opening a block explorer, and forming a judgment. The bear market changed the economics. When Terra collapsed in May 2022 and FTX followed in November, the surviving funds discovered that the only way to cover the long tail of protocols was to automate. A junior analyst at a mid-tier fund now reviews forty projects a week. No human reads forty whitepapers a week. So the reading got delegated to models, and the models got chained together.
The architecture is always the same. Stage one extracts facts: title, source, core thesis, a list of information points, the protocols named, the time-sensitivity. Stage two consumes those facts and produces the nine-dimension grid โ technical, tokenomic, market, ecological niche, regulatory, team and governance, risk, narrative, and supply-chain transmission. It is a beautiful design. It mirrors how a real desk actually thinks. And it has a single point of catastrophic failure: if stage one returns nothing, stage two has two choices. It can admit the void. Or it can fill it.
I have spent twenty-five years watching what happens when institutions choose to fill it.
The economics are unforgiving. A mid-tier fund that wants to stay in deal flow must review more projects than it can staff, so it buys coverage by the unit. Research vendors charge per report, which rewards throughput over accuracy. The fastest way to raise throughput is to shorten the human review step, and the fastest way to shorten that is to let the model write the conclusion as well as the summary. At that point the analyst is no longer analyzing. The analyst is editing, and editing is a far weaker check than investigating. I have run both kinds of desks. The difference in output quality is not marginal. It is categorical.
The bull market has accelerated this to a fever. Capital is flooding into anything with a credible ticker, and the demand for diligence has outrun the supply of diligence. Yields are not given; they are engineered โ and so is research. When a fund needs a memo to justify a position it has already decided to take, the memo is not an investigation. It is a rationalization with a table of contents. The pipeline does not care. It will produce the shape of rigor on command.
This is where my 2020 DeFi Summer experience becomes relevant. I deployed $200,000 across Compound and Uniswap pools and ran a dynamic rebalancing strategy that captured a 45% APY before the correction. I documented every position in real time. That documentation is the only reason I can now distinguish a genuine yield source from a subsidized one. Without the primary data, the report would have been indistinguishable from the marketing. That is the difference between an audit and a press release: the story is the asset; the code is the proof. When you have neither, you have an empty template.
Core: the mechanism of a hollow report
Let me dissect how a report becomes hollow, because the mechanism is more instructive than the outcome.
The two-stage design has a schema contract. Stage one promises to deliver an information-point list โ discrete, verifiable claims extracted from a source document. Stage two treats each of those claims as a fact-anchor and reasons outward. The whole edifice rests on the assumption that stage one's list is non-empty and its claims are true. Break that assumption and you get one of two failure modes.
Failure mode one is the visible one: the empty template. No facts arrive, the analysis refuses to invent, and the output is a skeleton with no tissue. This is, perversely, the safe failure. It wastes time. It does not cost money.
Failure mode two is the invisible one: the confident fabrication. A scrape returns partial text, or a field maps to the wrong column, or the source document is a marketing post that the extractor mistakes for a factual claim. Stage two receives a poisoned anchor and reasons flawlessly from a false premise. The output is fully populated, internally consistent, and wrong. It cites an unlock schedule that does not exist. It attributes a partnership that was never signed. It assigns a risk score to a team that has already dissolved.
I have seen failure mode two up close. In 2021, I led a ten-thousand-word investigation into the Bored Ape Yacht Club, interviewing fifty community leaders and clustering on-chain wallets to map the social hierarchy of early adopters. The piece predicted a shift from speculative trading to brand-equity building. What made it credible was not the prose. It was the fact that every claim traced back to a wallet, a timestamp, or a named interview. Culture is the only moat that cannot be forked โ but culture, too, has to be measured before it can be monetized, and measurement means primary data.
Strip the primary data out and you do not get a weaker version of that piece. You get its opposite: a document that flatters the reader with structure while concealing the absence of evidence.
The nine-dimension grid makes this worse, not better. Each dimension โ technical, tokenomic, market, ecological, regulatory, governance, risk, narrative, transmission โ is a place to hide. A risk matrix with six rows and four columns looks rigorous even when every cell says "N/A." A tokenomics table with a team allocation, a seed allocation, a community allocation, and an unlock schedule signals expertise even when the percentages are placeholders. The grid is a machine for manufacturing the appearance of coverage. We do not chase trends; we audit their foundations โ and the foundation of a grid is the data underneath it.
The regulatory dimension deserves special attention, because it is where the empty template does the most damage. A Howey analysis has four elements: investment of money, common enterprise, expectation of profit, and reliance on the efforts of others. Four rows, four cells. When all four read "N/A," the reader learns nothing โ but the reader also feels that a securities question was asked and answered, which is worse than never asking. In 2024, ahead of the spot Bitcoin ETF approvals, I wrote a strategic brief for major Brazilian pension funds, translating cryptographic security models into traditional fiduciary risk metrics. The entire value of that document lay in its specificity: custody arrangements, non-correlation assumptions, drawdown parameters, the exact regulatory perimeter in each jurisdiction. A pension committee does not allocate on a template. It allocates on a named custodian and a signed legal opinion. An empty Howey grid is not a conservative finding. It is a placeholder dressed as prudence.
The tokenomics dimension fails the same way. A supply table with rows for team, early investors, community, and treasury communicates diligence even when every percentage is unknown. But the unlock cliff is the single most predictive variable in a new listing, and it is almost always knowable โ it lives in the token contract, in the vesting schedule, in the on-chain claims. A report that marks "team allocation: N/A" when the contract is public is not being cautious. It is being lazy behind a screen of caution. I learned that lesson the hard way during DeFi Summer, when a rebalancing strategy that looked optimal on paper nearly came apart because I had modeled emissions as a constant when they were, in fact, a decaying curve with a cliff I had not read carefully. The data was on-chain the entire time. The report was only as good as my willingness to read it.
Consider what a genuine technical audit of that $100 million modular DA protocol would have surfaced. Data-availability sampling is a real architectural advance; my 2022 work on modular designs argued that fragmentation was the only viable path forward, and I quantified the cost-efficiency gains of sampling to convince skeptical institutional readers that the bear market was a pruning phase, not a funeral. But sampling does not eliminate the operator economics problem. Someone still pays for the proofs. ZK rollup proving costs remain absurd โ an operator running a prover without subsidized gas is bleeding money, and a bull market hides the wound because fee revenue and token price paper over the deficit. A real report would have said that. The empty template said "insufficient information," which is honest but useless. The fabricated report would have said "best-in-class proving efficiency," which is useful and false.
Sentiment is measurable, and that is precisely why its absence in a report is damning. Wallet clustering, funding-rate divergence, social-volume-to-price ratios, the ratio of new addresses to returning ones โ reading the silent language of digital tribes is a discipline, not a vibe. A report that leaves the sentiment dimension blank is not being humble. It is declining to do the work that distinguishes a durable community from a rented one.
Now scale the problem. One dossier is an anecdote. A thousand dossiers a week, generated across a dozen funds, is a market infrastructure. When the research layer hallucinates in unison, it does not merely mislead individual investors. It manufactures a consensus that has no empirical floor. Prices are set by marginal buyers, and marginal buyers increasingly act on machine-generated conviction. The feedback loop is short and self-reinforcing: a fabricated "partnership" appears in one report, gets cited by a second, gets aggregated by a dashboard, and within seventy-two hours it is treated as established fact by traders who never saw the original. There is no correction mechanism, because no one in the chain holds the primary source.
I watched this pattern in 2017, during the ICO era, in slower motion. Tokens issued without functioning contracts, described in whitepapers that promised roadmaps nobody audited. My team's reentrancy finding on that Waves module was unglamorous โ a state-ordering bug, not a headline โ but it forced a two-week delay and, more importantly, it forced a conversation about what the code actually did versus what the marketing claimed. That conversation is what a research pipeline is supposed to automate. What it has actually automated is the marketing.
Here is the part that should unsettle the desks relying on these pipelines. The extraction layer is not neutral. It inherits the biases of its sources. A project with a loud community generates more text, more posts, more scrapable surface area, and therefore more extractable "facts" than a quieter project with better code. The pipeline does not measure quality. It measures legibility. In a bull market, legibility correlates with price, so the research output inherits the market's own reflexivity: the projects that pump hardest get the densest reports, and the densest reports justify more buying. Dissecting the anatomy of a market illusion requires standing outside that loop. A pipeline that lives inside it can only amplify the illusion it was hired to audit.
The schema contract is where this becomes a governance problem rather than a technical one. When stage one and stage two disagree about the format of truth โ when the extractor emits an empty list and the analyst expects a populated one โ the system has no defined behavior. The rule I have always imposed on my own desks is simple: when the facts are absent, the report says so, loudly, at the top, in the largest type available. An honest void is worth more than a beautiful fabrication. But honesty is expensive when the client is paying for coverage and the bull market is paying for optimism.
Contrarian: the empty template is the honest artifact
The comfortable reading of all this is that the empty template is the scandal โ that somewhere, a machine failed and someone should be fired. I want to argue the opposite. The empty template is the only honest document in the stack.
Think about what it actually did. It received no facts. It refused to invent facts. It reported the absence. In a market saturated with confident, well-formatted, fabricated diligence, a document that admits it knows nothing is the closest thing to integrity the research layer has produced all quarter. The scandal is not the empty template. The scandal is the filled ones, and the fact that we cannot tell them apart.
This is the counter-intuitive core. We have built an industry that rewards the appearance of knowledge and punishes the admission of ignorance, and then we are shocked when the knowledge turns out to be appearance. The empty template violates an aesthetic norm โ it looks unfinished โ while satisfying an epistemic one: it is true. Every hallucinated report does the reverse. It looks finished and it is false. Given a choice between a document that is ugly and true and one that is elegant and false, the market consistently chooses the second, because the market prices presentation, not provenance.
I will go further. The obsession with complete coverage is itself the root cause. The nine-dimension grid exists because someone decided that every project must be evaluated on every axis, whether or not the data supports it. That is a template thinking error imported from traditional equity research, where coverage is mandatory because disclosure is mandatory. Crypto has no mandatory disclosure. Filling nine dimensions for a project with one scrapable blog post is not rigor. It is theater. The disciplined move is to evaluate the three dimensions where data exists and mark the other six as unknown โ and to defend that partiality rather than apologize for it. We do not chase trends; we audit their foundations โ and a foundation that has not been poured cannot be audited, no matter how many columns you draw for it.
Takeaway: the next narrative is provenance
The next narrative in this market will not be another layer, another chain, or another yield primitive. It will be provenance โ the verifiable lineage of a claim, from primary source to published conclusion. The story is the asset; the code is the proof, and the missing piece has always been the audit trail between them.
The protocols that survive the next cycle will be the ones that can show their work: on-chain evidence, timestamped interviews, reproducible metrics, and the intellectual honesty to mark what they do not know. The research desks that survive will be the ones that treat an empty template not as a failure to hide but as a signal to publish โ because the moment a pipeline can tell you why it knows nothing is the moment it can be trusted to tell you what it knows.
So here is the question I would put to every fund still running an automated diligence chain. When your model hands you a fully populated report on a project it cannot actually see, do you have the discipline to ask what it was looking at โ or do you have a position to justify before the round closes? The answer will determine which of us is still here when the euphoria ends, and which of us was just reading our own reflection in a well-formatted table.