The document that landed in my inbox last month had nine sections, forty-one assessment fields, and a cover page that read: PASS โ no material red flags identified. Every field beneath that cover page read N/A โ insufficient information.
The protocol in question had closed a nine-figure round eleven days earlier. Its testnet had processed fewer than four thousand transactions. Its token allocation table existed nowhere on the public internet. The memo was not a lie, exactly. It was worse. It was the flawless output of a pipeline that had failed at step one and had been engineered, by omission, to describe that failure as cleanliness.
I have watched this artifact circulate three times now, in three different funds, wearing three different logos. It is always the same document. A blank diligence template is indistinguishable from a clean audit to anyone who only reads the verdict line.
The pipeline nobody audits
The modern crypto research stack has two stages, and only one of them is ever tested.
Stage one is extraction: a scraper pulls the source article, a parser breaks it into atomic facts โ funding amount, chain, auditor, unlock schedule, jurisdiction โ each anchored to a quotable line in the original text. Stage two is scoring: those facts get mapped onto a framework and graded. Technical. Tokenomics. Market. Ecosystem. Regulatory. Team and governance. Risk. Narrative. Supply-chain transmission.
Stage two gets all the attention. It gets the beautiful dashboards, the partner-facing PDFs, the conference panels about rigorous frameworks. Stage one gets a cron job and an intern.
So when the article sits behind a paywall, or renders client-side through a JavaScript bundle that returns an empty div to a headless browser, or trips an anti-scraping rule, or arrives with a mangled character encoding โ stage one returns nothing. And stage two, being a template rather than a mind, does not error out. It emits.
What it emits looks like analysis. Nine sections. Forty-one fields. A risk matrix with six categories and five columns each. It is, structurally, indistinguishable from a real report โ the same way a shell company is structurally indistinguishable from a company.
This is where I have to be blunt about my own history, because I did not arrive at this skepticism from a whitepaper. I arrived at it in 2017, at twenty-one, holding three altcoins and a naive belief that code is law, one of which rug-pulled in under a week. I spent the 2018 bear market not doing technical forensics but doing autopsy: why did these things die? The answer was almost never the code. It was that the token model had no revenue and the diligence had no facts. From whitepaper fantasy to ledger reality, the gap is always a data gap, never an opinion gap.
And here is the uncomfortable part. The document with forty-one N/A fields is the most honest thing produced by that pipeline all quarter. It is the only page in the stack that admits what it does not know. Every other output has been smoothed into prose confident enough that no allocator will ask where the numbers came from. When the algo breaks, the axiom remains: an unverified claim and a fabricated claim are the same claim in a portfolio construction model.
The economics of not checking
Why does this failure mode persist in a market this liquid? Because in a bull market, the marginal revenue of shipping a research note exceeds the marginal cost of verifying it โ and verification is the only line item you can cut without anyone noticing.
Run the arithmetic from my own audit work. A genuine single-protocol deep dive โ contract review, allocation tracing across foundation and team wallets, chain-level flow reconstruction, jurisdiction mapping โ costs somewhere between forty and a hundred and twenty analyst hours. Multiplied across a book of thirty positions, that is a headcount, not a budget line. A templated output costs minutes. The subscription price the allocator pays is identical either way, because the allocator is not buying facts. The allocator is buying a document that lets them defend a decision to an investment committee.
That is the whole mechanism. The product being sold is not diligence. It is audit-trail-shaped paperwork. And paperwork has one defining property: it only needs to be complete in form.
This is the same structural failure I have documented in decentralized governance for years. A DAO that exists as a multisig with an off-chain vote and a Snapshot page has the form of an organization and none of the legal substance โ and when something breaks, the form provides no shield at all, only the appearance of one. The nine-dimension report is the DAO of research artifacts. Same shape. Same comfort. Same absence underneath.
And it is the same failure I keep flagging in the data-availability conversation, where an entire category of rollups has been architected around dedicated DA capacity for data volumes that the aggregate user base will not produce this decade. We have built extraordinarily sophisticated infrastructure for receiving data, and almost none for confirming that data arrived. A blob with nothing in it does not fail. It posts. It gets attested. It settles. The empty input is not an edge case in this architecture; it is the honor system wearing a cryptographic costume.
The risk matrix that prices silence at zero
Now the part that actually costs money.
In February 2022 I built a correlation stress model for a client who wanted exposure to a large algorithmic stablecoin ecosystem. The model was crude by today's standards but it had one honest row: what happens to the collateral when the collateral's own price is the thing that is failing? The answer was a death spiral, and I said so, and I was told โ in a room, by a man with a much larger title than mine โ that my concern was overstated. The word he used was not "wrong." The word he used was "emotional."
Eleven weeks later the mechanism did exactly what the arithmetic said it would.
Here is what I took from that, and what it has to do with a memo full of N/A: an unassessed risk is not an unpriced risk. It is a risk priced at zero. The order book does not contain a row for "unknown." It contains a row for "clean." Absence of data gets read as absence of danger, because the alternative โ that the analysis is blind โ is a sentence no one wants to put in a committee memo.
So the blank template is not neutral. It is directional. It is a long position in the narrative.
Watch how it propagates. Stage one fails on a token unlock disclosure that lives in a PDF behind a login. Stage two writes N/A in the supply-structure table. The allocator reads a report with no red flags. The position sizes up. And then the unlock lands โ 14% of float, twelve months post-listing, exactly as the inaccessible document said it would โ and the desk calls it an unpredictable event.
It was not unpredictable. It was unparsed. The market doesn't price what it can't parse, and then it calls the bill a surprise.
I saw the same reflex in 2020, in the middle of DeFi summer, when I was a junior analyst watching peers chase triple-digit APYs and I kept asking where the yield actually came from. Almost none of it came from revenue. It came from the next depositor. When I published a thread arguing that the sustainable version of that structure required Bitcoin dominance to stay above a threshold most people considered irrelevant, the response was that I was conflating macro with micro. Two months later the correlation showed up in the data. The protocol-level metric had been nearly useless. The macro-liquidity metric had been the whole story.
That experience is why every framework I build now begins with the global liquidity map and only then descends into protocol specifics. It is also why I trust an N/A over an estimate. An estimate with a confident font is the most dangerous object in this industry.
What verification actually costs
I want to be precise about the minimum viable input, because "do more diligence" is the empty advice of our profession and I refuse to add to it.
One fact is enough to begin. Not a framework โ a fact. A funding amount with a named source. A treasury address with a matching on-chain balance. An unlock table with a date and a percentage. Five to fifteen atomic facts, each one anchored to a line someone can click, will outperform a nine-section scored report every single time, because the facts can be falsified and the scores cannot.
When I led the custodial-risk work after the spot ETF approvals in early 2024, the useful output was not a grade. It was a diagram of key custody: how many signers, in how many jurisdictions, under what legal structure, with what recovery path if two of them stop answering. That diagram took three weeks. It could not be templated. It also could not have been produced by any pipeline that failed at extraction, because the whole exercise consisted of finding documents that were not indexed, not publicized, and in two cases not in English.
That is the real cost of the empty input: not the wasted memo, but the argument that never gets made because the data never arrived to make it. Skepticism is the highest form of due diligence, but skepticism requires a surface to push against. You cannot doubt a blank field.
The contrarian read: do not fix the scraper
The instinct in this room will be to fix stage one. Patch the headless browser, rotate the proxies, add the OCR fallback. That is correct engineering and it is the wrong conclusion.
A pipeline that fails silently is a more informative object of study than the protocol it was supposed to analyze. The N/A document is not a report on the protocol. It is a report on the vendor. It tells you, with a candor no sales call would ever offer, that this firm's revenue depends on emitting something and only its integrity depends on emitting truth โ and that when those two conflicted, the firm chose revenue by default, without anyone in the loop having to make a conscious decision.
The firms that survive the next drawdown will be the ones whose pipelines fail loudly. An error page is worth more than a passing grade. Every allocator I respect now asks a question that would have seemed absurd in 2021: show me the extraction log. Show me which fields came back empty and what your system did when they did.
In a market where capital is arriving faster than verification capacity, the bottleneck is not analysis. It is the willingness to return an error. We don't need better templates. We don't need another scoring dimension. We need systems that would rather break than bluff.
Positioning the cycle
The signal to watch over the next two quarters is not price. It is what I have started calling N/A density โ the share of fields in a research corpus that resolve to insufficient information rather than a sourced fact. When N/A density rises across a sector while valuations rise alongside it, the gap is being filled by narrative, and narrative has a known half-life.
The trade is not to short the protocol. The trade is to underweight anything whose bull case cannot survive an extraction log, and to overweight the unglamorous layer that keeps receipts โ the indexers, the on-chain attestation services, the compliance shops that charge by the confirmed fact rather than by the page.
When the algo breaks, the axiom remains. The question for this cycle is simple and slightly uncomfortable: when your research stack returns nothing, does it tell you โ or does it hand you a pass?