N/A Is Not a Rating: The Null-Set Problem in Crypto Research

The document arrived formatted as a report. Nine analytical dimensions. Forty-one table cells. A risk matrix with six rows and six columns. A competitive landscape section with two placeholder rows.

Every field returned the same value: N/A.

I read it twice. The second pass was not for content — there was none — but for structure. The form had survived the total absence of information. Headings held their weight. Tables still aligned. The risk matrix still looked like a risk matrix. A reader skimming on a phone would have registered the silhouette of diligence and moved on to the next tab.

That is the anomaly. Not the blanks. The fact that the blanks rendered as a report.

In eleven years of writing on-chain forensics, I have learned that the most dangerous documents in this industry are not the ones that lie. Lies leave residue. You can trace a lie back to the wallet that funded it. The dangerous documents are the ones that are structurally complete and substantively empty, because they consume the reader's attention without ever returning a signal, and they do it while looking exactly like work.

So I did what I always do with an outlier. I stopped looking at the value and started looking at the instrument that produced it.

Context: How Research Became a Coverage Product

To understand why a blank template is a market event and not an editorial embarrassment, you have to understand what research in this industry actually is now.

Between 2017 and 2021, crypto research was a scarcity business. A handful of desks — Messari, Delphi, a few bank side-arms, and a long tail of independent writers like me — produced the only structured analysis that existed. The bottleneck was analytical labor. If you wanted a view on a protocol, you paid for someone's attention, and attention was expensive.

That inverted. By 2024, the bottleneck moved from analysis to coverage. Every fund with more than $50 million under management now maintains a universe list of 200 to 600 assets, and internal compliance requires a written memo on file for each one before a position can be opened. Not because the analyst has something to say. Because the file needs a document in it.

This is a well-understood pathology in traditional research and it has been imported wholesale. The memo becomes a compliance artifact rather than a knowledge artifact. Its function is to prove that someone looked, not to record what they saw. And once the memo's purpose is evidentiary rather than epistemic, the cheapest way to produce it is a template.

The nine-dimension framework I was handed is a good example of the genre. Technical. Tokenomics. Market. Ecosystem. Regulatory. Team and governance. Risk. Narrative. Value-chain transmission. It is comprehensive in the way that a checklist is comprehensive: it enumerates everything that could be asked, and in doing so it guarantees that most of what gets answered is irrelevant.

I have my own history with this failure mode. In 2017, at thirty-six, I ignored the ICO calendar entirely and spent six weeks building a Python simulation of the 0x relay incentive structure, because I wanted to know whether the fee distribution model actually paid relayers for the work they performed. It did not, quite. The flaw was theoretical and small and nobody cared, but the methodology stuck: a claim without a reproducible artifact behind it is a rumor with better typography. Since then I have refused to publish a view I could not reconstruct from the ledger.

That is why the blank template interests me. It is the purest specimen of the opposite practice — a claim-shaped object with no claim inside it — and it appears in this market thousands of times a day.

Core: Field Selection Is the Analysis

Here is the part that most readers of research never see. The choice of fields is the analysis. Everything downstream is arithmetic.

A template that asks for TVL cannot find a balance-sheet fraud, because balance-sheet fraud lives off-chain. A template that asks for APR cannot find a yield that does not exist, because the advertised APR is precisely the number the fraud controls. A template that asks for DAU cannot find wash trading, because wash trades are real transactions from real wallets and they inflate DAU honestly.

The instrument determines the discovery. A telescope pointed at the visible spectrum will never find a black hole; it will find a blank patch of sky and file it as N/A. The blank patch is not evidence of absence. It is evidence that the instrument was pointed at the wrong wavelength.

I want to walk through four cases from my own files where the standard field set returned a confident, well-formatted, and completely wrong answer.

Case One: The Ledger That Had No Token

In late 2022 I spent four months reconstructing FTX's collateral movements on Solana. I mapped a chain of roughly fifteen thousand transactions showing customer deposits routed into Alameda-controlled addresses and cycled back as collateral for positions that could not be liquidated without revealing the hole.

Now put that through the nine-dimension template.

Technical positioning: N/A — FTX was not a protocol, it was a custodian. Tokenomics: N/A — FTT existed but the failure was not in FTT's supply schedule; the failure was in an off-chain balance sheet that no token field could capture. Ecosystem dependencies: the template asks for upstream and downstream, and FTX's dependency graph was a single node with an arrow pointing at itself. Risk matrix: the six categories were technical, market, operational, regulatory, competitive, narrative. Custodial commingling was none of them.

The template would have returned a full page of N/A on the largest fraud in the industry's history, six months before the fraud became public.

This is not a hypothetical. It is what actually happened across the coverage universe. Dozens of desks held FTX equity and FTT and produced memos that scored it well, because the fields they measured — market share, revenue growth, brand — were all genuinely excellent right up to the moment they were worth nothing. The template was not blind to the risk. The template did not have a field for the risk, which is a different and more permanent condition.

What the instrument needed was not a new category. It needed a question: can I reconcile the entity's stated liabilities to an on-chain or audited asset base, and if not, what is the counterparty chain that guarantees them? One question. It would have taken four months to answer, which is exactly why nobody put it in the template.

Case Two: The Yield That Wasn't There

In the summer of 2020 I became fixated on Curve's stablecoin pools, which were advertising yields in the high double digits and were, at the time, the closest thing the market had to a risk-free rate.

I built a spreadsheet modeling five hundred liquidity scenarios. I isolated CRV emissions, decay schedules, and slippage under various depth assumptions. The conclusion was that the real yield to a liquidity provider was roughly eighteen points below the advertised figure once emissions decay and hidden slippage were priced in.

The template's APR field would have captured the headline. It would have captured it perfectly, in a clean table cell, with a date stamp. It would have been accurate and useless.

The number that mattered was not the APR. It was the derivative of the APR with respect to time, weighted by the depth of the pool at each point. No template asks for that, because it requires modeling rather than lookup, and modeling does not scale to a six-hundred-asset universe.

This is the structural trade-off nobody states plainly. Coverage and depth are inversely proportional, and the industry has chosen coverage while marketing it as depth. The spreadsheet I built took nine days. Nine days times six hundred assets is sixteen years of labor. So the market bought six hundred lookup tables instead, and the lookup tables were right about the numbers and wrong about the money.

Case Three: The Volume That Never Existed

In 2021, while the market was pricing JPEGs, I pulled the on-chain transaction record for a major blue-chip collection and matched wallet pairs against overlapping transaction histories — same funding source, same gas price pattern, same inter-transaction intervals, self-matched trades cycling through two to four addresses.

Roughly sixty percent of floor-price movement was coming from those pairs. The true market depth was about a fifth of reported volume.

Here the template's fields were not merely insufficient. They were actively misleading. Daily active users: inflated, and honestly so. Retention: excellent, because the same wallets came back every day. Volume: enormous. Market share: dominant. Every cell green.

The instrument had no field for self-matched trade, because self-matched trades are not a category anyone thought to exclude — they are just transactions, and transactions are the unit of truth in this industry, and that is the whole problem. On-chain data is not self-authenticating. It is self-documenting. The distinction is the entire job.

Following the trail of outliers that others ignore means, in practice, following the wallets that appear too often, too regularly, and too cheaply. The bots were not hiding. They were generating exactly the data the template rewarded.

Case Four: The Inflow That Inverted

After the spot Bitcoin ETF approval in January 2024, I started tracking daily creation and redemption data for the largest issuer. The consensus field set for that product was: net flows, assets under management, fee, and market sentiment.

What the data showed was a negative correlation between large single-day inflow prints and short-term price performance over the following three to ten sessions. High inflow days preceded corrections with uncomfortable regularity, because the marginal buyer on those days was an arbitrageur completing a basis trade, not an allocator expressing a view. The inflow was the exit liquidity for the trade, not the entry of conviction capital.

I published a model on that basis and it flagged the March 2024 drawdown before it happened.

Now run it through the template. Sentiment: greed. Flows: strongly positive. Institutional adoption: accelerating. Risk: low. The template would have generated a maximally bullish scorecard at precisely the moment the marginal buyer was a seller in disguise.

The algorithm does not lie, but it may omit. Flows were positive. That was true. What was omitted was the identity of the counterparty on the other side of the flow, and identity is not a field.

The Null Set Is a Position

The report I was handed is careful to say that N/A does not equal low risk. That is correct, and it is also the most important sentence in the document, and it is buried in a footnote-shaped paragraph.

Here is the stronger version. A document that returns N/A across every field is not neutral. It is a claim that nothing is knowable — and in a market where someone always knows something, that claim is itself a tradeable signal, and the direction of the trade is against you.

Think about what an all-N/A report actually communicates to the person who commissioned it. It says: the instrument found nothing. The reader's rational response is to treat the asset as unremarkable and allocate attention elsewhere. But the correct response is the opposite. An all-N/A return means the instrument failed to map onto the object, which means the object has properties outside the measurement frame, which is precisely where the mispricing lives.

There is a second-order version of this that I find more interesting. Consider three of the field gaps I have already described, mapped onto three positions I hold.

First, public-goods funding. The template has no field for attribution — for whether a grant produced a public good that would not otherwise have existed. It has fields for spend and for governance participation. Optimism's RetroPGF is the only mechanism I have seen that measures the counterfactual rather than the expenditure, and a template built around spend and turnout would score it as bureaucratic overhead while scoring a committee-driven grant program as efficient. The instrument rewards the wrong thing, consistently, and then reports the result as data.

Second, DEX architecture. The template has a field for features and no field for developer attrition. Uniswap V4's hooks make the pool a programmable primitive, and the complexity curve that comes with that is steep enough that most teams will not climb it — not because the design is wrong, but because the number of teams capable of shipping a correct hook, auditing it, and maintaining it across upgrades is small and does not grow on the timeline the market expects. A feature checkbox cannot see this. An attrition curve can.

Third, rollups. The template asks for transactions per second and total value locked. It does not ask for proving cost per batch divided by fee revenue per batch. That ratio is the entire economic story for ZK rollups right now, and at current proving costs against current gas prices it does not clear. A template measuring throughput would report a healthy network. The network is healthy. The operator is bleeding.

Three positions, one mechanism: the field set was designed around what is easy to look up, and the truth is located in what is expensive to derive. That is not a bug in one template. It is the equilibrium of the research market.

Contrarian: In Defense of the Honest Blank

I want to argue against myself for a moment, because the case is stronger than it looks.

Consider the alternative to the blank template. It is not a rigorous nine-dimension analysis. It is a filled-in template — the same structure, with plausible numbers in every cell, generated in twenty minutes by someone who has read the project's documentation and nothing else. That document is worse than the blank one in every dimension that matters. It is worse because it is unfalsifiable at a glance, because it launders narrative into the visual grammar of analysis, and because it will be cited downstream by someone who never checks.

The blank template is at least honest. It says: I do not have the inputs. That refusal is rare, it is correct, and in a market where ninety percent of published research is a filled-in form, the honest blank is the more valuable artifact. I would rather read a page of N/A than a page of confident fiction, and I have read a great deal of both.

But I do not accept the framing, and here is where I break with it. An honest blank is only valuable if it triggers a data-acquisition action. If it does not, it is a well-formatted refusal — a document that discharges the analyst's obligation while leaving the reader exactly as uninformed as before, which is the compliance-artifact problem wearing a different coat. The N/A is a starting gun, not a finish line. A report that says "information insufficient, therefore no assessment" and stops has done the easy half of the work and billed for the whole thing.

The correct output of an all-N/A pass is not a document. It is a work order: five specific facts to acquire, with named sources, owners, and deadlines, and a stated position on what you will do if the facts do not arrive by Friday. The report I read had none of that. It had a required-inputs appendix, which is close, but an appendix is where findings go to be ignored.

So the contrarian claim is narrower than it first appears. The blank is better than the lie. But the blank is not the answer; it is the question, and a question with no follow-up is just a longer silence.

Takeaway

The signal I will be watching over the next several weeks is not a price level or a flows print. It is whether the research shops that publish coverage universes start publishing their field dictionaries alongside their conclusions.

If a desk will not tell you which fields it measures, you cannot know what it can and cannot see, which means you cannot distinguish a genuine N/A from a structural blind spot. When you start seeing field dictionaries — and someone will do it, because the credibility arbitrage is enormous — the first thing to check is whether the dictionary contains a field for counterparty identity, a field for attribution, and a field for cost per unit of work. If those three are missing, the report is a coverage product regardless of how many dimensions it claims.

And if your own desk is still running nine-dimension templates that return N/A on assets nobody has explained, do not file the report. The blank cell is not the finding. The blank cell is the map of where you have not looked.

Deciphering the hidden geometry of liquidity pools is easy compared to this. The pools at least want to be measured. A template is designed not to be.