Hook
Last week a 4,000-word research document crossed my desk. Nine sections. A risk matrix. A Howey test table. A supply-structure breakdown. A star rating for "information value." Every substantive field carried the same three words: N/A — insufficient information. No token. No chain. No team. No numbers. The document did not know what it was about, and it graded its own findings anyway.
That is the artifact worth dissecting. Not the missing data — missing data is the normal condition of crypto research; most of it is written before the facts arrive. The anomaly is that the scaffold survived the void. A schema populated entirely with nulls still shipped with priority-ranked warnings, a compliance section, and "opportunity identification." Formatting did the work that evidence was supposed to do.
I have read thousands of these documents. They are the most under-audited instrument in this market. Nobody checks them, because they look checked.
Context
Here is the production line. A research pipeline has two stages: extraction and analysis. Extraction pulls facts from a source — title, claims, project name, numbers. Analysis runs those facts through a fixed framework. When extraction returns empty, a well-behaved system should halt and flag the failure. It almost never does. The schema is hard-coded, so the second stage executes anyway, filling every row with its default value.
The result is a document that is complete in form and empty in substance. Every category has a row. Every row has an entry. The entries are "N/A." Research desks, alpha groups, and newsletter farms now run this exact pipeline at scale, because the marginal cost of one more formatted page is nothing and the marginal perceived value is high.
The schema itself is not arbitrary. Most of these frameworks were written by people who lived through specific failures — the Terra collapse taught everyone to check for a credible backstop, the Curve reentrancy taught everyone to read the invariant. The fields are scars. But a field written after a disaster is a field that must always be filled, and a system that must always fill a row will fill it with whatever is available, including nothing.
There is a precise software analogy, and it is the reason I distrust these outputs. In most languages an unset variable is loud — null, undefined, a runtime error that halts execution. In Solidity, an uninitialized storage slot is not null. It is zero. It is a valid value of the correct type. The contract does not complain. It returns 0 and continues.
That is exactly what a default-filled framework does. It returns zeros and prints them as findings. A category is not an answer. Enumeration is not examination. The framework substituted one for the other, silently, and the output looked like work.
Core
Three mechanisms turn an empty schema into apparent diligence.
The first is category substitution. "Security assumptions" is a heading, not a conclusion. But once it sits in a table with an adjacent cell, the reader's eye closes the gap. The heading becomes the finding.
The second is symmetry bias. Because every framework category gets a row, every category appears to have been considered. This is false. A list of questions is not evidence that the questions were asked. But a table is expensive-looking, and we have been trained to read expense as effort.
The third is rating drift. The document I opened carried a five-point "information value" scale and awarded its own inputs one star. One star reads as "low quality" — a judgment about the asset. It is not. It is a judgment about the analyst's empty input. Readers collapse the two, and a document that knows nothing becomes a document that says "caution," which is indistinguishable, at a glance, from a document that knows something.
I learned how load-bearing these gaps are in 2020, reverse-engineering the Curve stableswap invariant. The whitepaper under-specified the mechanism. That silence was not neutral. The slippage exploit lived precisely where the documentation stopped talking. I audited the void and found a backdoor. Under-specification is not the absence of risk. It is where risk accumulates, unobserved, until volatility prices it.
I met the same wall in 2021, running statistical clustering on NFT floor data. My model ranked assets by trait rarity and sales velocity, and it worked — until it didn't. Three positions I could not exit, because the model scored value and ignored depth. The data was real. The liquidity was not there. A framework can be internally consistent and still be describing a market that does not exist.
The same holds for an empty analysis field. "N/A" is not "unknown." It is "unmeasured." Unmeasured risk does not sit still — it compounds while the report sits in a folder looking finished. Floor sweeps are just data points in motion, and so is every blank cell: it is a position someone is holding without a price on it.
Contrarian
The counter-intuitive claim: bad data is safer than structured missing data.
Give the pipeline its due: an N/A is more honest than a hallucinated number. A model that invents a $40M TVL figure is actively dangerous; one that writes "insufficient information" is not lying. That is the strongest defense, and it fails. Honesty about the input becomes authority about the output the moment it is formatted. The disclaimer is buried on page nine. The risk matrix is on page one.
When a report is obviously garbage — a shill thread, a paid promotion — you discard it. When a report is formatted correctly, you file it. You cite it. Structure is a trust heuristic, and it was a reliable one for decades precisely because producing a professional risk matrix was expensive. That cost has collapsed. The marginal cost of generating a nine-section, properly formatted, entirely vacuous due-diligence document is now approximately zero. The cost of reading one as evidence has not adjusted at all.
The rating is worse than the emptiness. A star scale applied to absent information is not a neutral gesture. It asserts that the axis is real, that a five-star version of this document exists somewhere. It invites you to compare nothing to nothing and call the result research.
This is not a failure unique to automated pipelines. It is the dominant pattern of the last three years of RWA decks — regulatory boxes checked, settlement rails absent, the template complete and the infrastructure not. The document was always the product. Smart contracts execute truth, not intent, and a schema executes its own structure whether or not there is a subject inside it.
Takeaway
The filter that scales is not better models. It is a single question: name one field in this analysis you could be wrong about.
If every field is N/A, there is nothing to be wrong about — which means there is nothing to be right about either. Falsifiability is the only due-diligence standard that survives contact with a market that can now generate confident-looking structure for free. The next edge is not in reading more reports. It is in refusing to consume the ones that were never about anything.
