Nine dimensions. Zero data points.
Last week I ran a staged research pipeline against a crypto asset. The architecture is standard in my workflow now. Stage one deconstructs the source into facts: title, publication date, information points, identified protocols, source quality. Stage two scores nine dimensions โ technical architecture, token economics, market positioning, ecosystem role, regulatory exposure, team and governance, risk surface, narrative versus delivery, and supply-chain transmission.
Stage one returned empty. Not sparse. Empty. No title. No date. No information points. No protocol named.
The report that came out the other side refused to fill the gaps. It printed N/A forty-one times, flagged the empty payload as a blocking defect, and closed with three remediation paths. That refusal is the most useful document I have read this quarter, and it concerns an asset I still cannot identify.
Here is the part that should unsettle you. I have watched this exact failure mode manufacture bullish theses.
The industrialisation of diligence
Research in this market has been industrialised, and industrialised processes fail in industrialised ways. Two years ago a deep dive meant an analyst, a forum thread, and a docs page. Today it means a chain: scrape, deconstruct, score. Each stage hands off to the next. Each handoff is a place where an empty payload can travel downstream wearing the shape of a legitimate input that merely happened to be thin.
The triggers are mundane. A paywalled source. A body rendered by JavaScript the crawler never executes. A link that resolved to a redirect and returned an empty string instead of an error. An article published as an image. A model call that timed out and returned well-formed JSON with nulls, because emitting valid-shaped nulls is cheaper than raising an exception.
None of that is exotic. What is exotic is the response. The default behaviour of a scoring system is to score. Hand a nine-dimension rubric an empty input and it will not raise its hand. It will interpolate. Technical innovation: moderate. Token distribution: typical for the sector. Regulatory risk: uncertain. Those three phrases are not analysis. They are the sound of a model covering for a missing file.
In 2022 I spent the drawdown auditing lending-pool contracts, and the lesson that transferred was not the reentrancy bug I found in a withdrawal function. It was the default. A function that returns success on a failed state check is not a convenience. It is a lie with a clean interface. Research pipelines carry the same bug class. An empty stage-one output that returns HTTP 200 is a failed state check wearing a green status code.
Regulation has now priced that bug. Under MiCA, an investment memo that cannot name a jurisdiction, an issuer, and a publication date is not merely sloppy. It is undocumented. Compliance costs I modelled in 2025 put annual legal overhead for a small Stockholm-based rollup near โฌ150,000 โ and a meaningful share of that is the documentation of exactly the fields this pipeline dropped.
Every dimension has a data prerequisite
The nine-dimension template is not the problem. The template is correct. The problem is that most operators treat the dimensions as questions rather than as functions of inputs. Each one has a minimum data prerequisite. Strip the input and the dimension does not become neutral. It becomes undefined.
Token economics is the clearest case. Supply schedule, allocation across team and early investors, unlock cadence, emissions rate, revenue routing โ remove any of those and the analysis is not preliminary. It is fiction. A distribution table with N/A in every row is not a conservative estimate. It is a blank page with gridlines.
Regulatory exposure is second. The Howey factors are not a vibe. They are four factual predicates โ capital contribution, common enterprise, expectation of profit, reliance on others' effort โ and each requires specific evidence: who received the tokens, at what price, under what lockup, with what promises attached. With no allocation data, the securities determination is not uncertain. It is unavailable.
Market positioning is third, and here the failure is temporal. The single most damaging missing field in the entire report was the publication date. Without a timestamp, pricing-in is unmeasurable. You cannot distinguish a rumour from a confirmation, a leak from a retrospective. Every market-structure judgement collapses into present tense. This is why I now treat a missing date as a hard stop rather than a soft flag โ it invalidates an entire dimension, not one cell.
The dimensions that fail silently
Two of the nine degrade without announcing it. Ecosystem role and supply-chain transmission.
Ecosystem role requires upstream and downstream relationships: which protocols depend on this one, which depend on it in turn, where developer activity concentrates. Remove the project name and the dimension still produces text โ generic text about composability and network effects. That text is worse than N/A, because it survives review. A reviewer scanning for blanks catches N/A in two seconds. A reviewer scanning prose catches nothing.
Supply-chain transmission behaves the same way. Miners, exchanges, infrastructure, DeFi, NFT and GameFi, TradFi โ six nodes, six impact directions, six timeframes. With no identified counterparties, the honest output is a row of blanks. The dishonest output is positive for infrastructure, neutral for DeFi, which is what a model writes when it has learned that this is the shape of a plausible answer.
This is the practical argument for machine-checkable nulls. Make N/A a required enum value rather than a prose absence. Then count them. If a report returns more than a threshold of nulls โ say, four of nine dimensions โ the pipeline should not emit a score at all, regardless of how confident the remaining five look.
Abstention is a position
Here is where I diverge from the consensus on research quality. Most analysts treat insufficient data as a gap to be closed later. I treat it as a market state with its own information content.
Yields attract capital, but security retains it. The same asymmetry governs information. A protocol that publishes a supply schedule retains analytical capital; one that does not attracts only narrative. The absence of verifiable fundamentals is not a neutral observation. In a sideways tape it is a distribution signature. Assets with real cash flow and real documentation keep their liquidity through the chop. Assets that survive on vibes lose it quietly, week by week, while the price chart stays flat and the order book thins.
I built a Security Risk Score into my reports for the same reason I now want a Data Integrity Score. The logic is identical. A protocol's market cap tells you what the market paid. It tells you nothing about whether the thing underneath will still be there in eighteen months. Code integrity is one axis. Disclosure integrity is the other. Score both, or you are ranking assets on the quality of their marketing.
The five-field minimum
If you run any version of this pipeline, here is the gate I would install. Five fields, all mandatory, all validated before stage two executes.
Title and source. Publication date, parsed as an absolute timestamp, not recent. At least five information points, each with an identifiable subject and a fact. A named protocol or issuer, or an explicit none identified. And a source-quality flag distinguishing primary disclosure from secondary reporting.
If any of the five fails, the pipeline halts and returns a null report. Not a partial score. A null report, with the defect named and the remediation path attached. The cost of halting is one delayed memo. The cost of not halting is a โฌ50M allocation decision resting on a table of interpolated adjectives.
I learned the size of that cost in 2024. Building the liquidity model that correlated Federal Reserve balance-sheet expansion with the ETH/BTC pair, I had to reconcile roughly โฌ50M of institutional inflow data against a narrative that said ETF approval was sufficient on its own. It was not. Flows without M2 expansion produced drift, not repricing. That conclusion was only reachable because every input was dated, sourced, and auditable. Had a single stage returned nulls and been quietly filled, I would have published a confident, wrong thesis with a clean chart attached.
From the lab experiment to the global standard โ that is the arc I keep returning to. Backtesting stablecoin peg stability with โฌ5,000 of my own savings in 2020 was a lab experiment. It was allowed to be messy. Institutional research is not. It is the global standard, or it is nothing, because the moment it scales it starts informing allocations that move real balance sheets.
The 2026 problem: machines writing diligence about machines
There is a second-order failure arriving, and it worries me more than empty payloads.
In 2026 I evaluated the data availability layer of autonomous AI agents settling proofs through decentralised storage. The finding that stuck: only 12% of the agents I modelled could sustainably fund on-chain verification of their own output. The rest generated content, claims, and reports with no economic mechanism to attest that any of it was true.
Apply that to research. We are entering a cycle where the supply of analysis expands faster than the supply of verifiable data. Agents will produce nine-dimension reports on assets whose stage-one inputs were empty, and they will do it at zero marginal cost, with perfect formatting, and no instinct to abstain. The AI liquidity trap I described for compute markets has an information analogue: without tokenised attestation of provenance, machine-generated diligence becomes isolated from the economics it claims to describe.
Which means the human edge is no longer synthesis. Synthesis is commoditised. The edge is refusal โ the discipline to return a null report when the payload is empty, and the infrastructure to detect emptiness before it is scored.
The contrarian read
The market does not pay for abstention. It pays for conviction, delivered on schedule, in a format that fits a slide. That incentive structure is precisely why the null report is rare and why it is valuable.
But I will not romanticise it. Most N/A outputs are not integrity. They are routing bugs. The difference matters, and it is testable. An honest null report names the defect, identifies the failing stage, and specifies the remediation. A lazy one shrugs. The report I received did the former: it flagged the empty stage-one payload as a blocking condition, refused to fabricate technical assessments, and refused to assign a risk grade to a protocol it could not name. That is a control, not a mood.
The blind spot in the opposite direction is worse. Analysts who never abstain do not produce more coverage. They produce coverage of a fictional asset universe, where every protocol has a moderate risk profile and a typical unlock schedule. That universe is comfortable and it is expensive. The cost is paid later, in the drawdown, by whoever trusted the table.
The uncomfortable corollary: in a consolidation market, the assets with the thinnest verifiable disclosure are also the ones with the thinnest bids. The N/A is not only an analytical state. It is, more often than comfortable, a market one.
Positioning for the next cycle
So here is where I land, and it is not a summary.
The next twelve months of this sideways tape will be decided less by who identifies the winning narrative than by who can distinguish a documented asset from an undocumented one. Liquidity will keep rotating into anything with a chart that moves. It will stay in the things that can survive an audit of their own disclosure.
Build the gate before you need it. Five fields, mandatory. A null report as a valid output. A Data Integrity Score sitting next to the Security Risk Score on every memo you write, because a protocol that cannot document its own supply schedule is telling you something about how it will document its next incident.
Yields attract capital. Security retains it. Disclosure is what makes the difference legible.
The question I keep returning to is not which asset the empty report described. It is how many reports you have read this quarter that were also empty โ and how many of them scored well anyway.