On the morning of March 3, a term sheet crossed my desk with a hundred-million-dollar headline and eleven empty fields.
The headline was the round. A freshly funded infrastructure protocol, Series B, $100 million at a $1.4 billion post-money valuation, led by a fund whose name appears in every crypto LP deck published since 2021. The eleven empty fields were everything else. Token supply schedule: not disclosed. Vesting cliff: not disclosed. Unlock cadence for the insider tranche: not disclosed. Auditor identity: not disclosed. Contract upgrade authority: not disclosed. Revenue, if any: not disclosed. Jurisdiction of incorporation: not disclosed. Founder identity beyond two LinkedIn profiles: not disclosed. Treasury runway: not disclosed. Governance concentration at genesis: not disclosed. And the eleventh field, the one that should have stopped the meeting cold โ the identity of the entity that will custody the raise.
The market, of course, priced the headline. Within ninety minutes of the leak, the associated tickers on three venues moved between four and nine percent, and a derivative listed on a mid-tier exchange printed open interest that had not existed the previous week. Nobody priced the blanks. That is the anomaly I want to dissect, because it is not unique to this deal, this quarter, or this cycle. It is the defining microstructure of crypto asset research in 2026, and it is where the last genuinely mispriced risk in the asset class now lives.
I. The Shape of the Problem
I have been producing institutional crypto research for a little over two decades, first as an observer of the 2003-2008 commodity complex, then as an analyst of the 2017 ICO era, and now as a risk assessor at a Zurich desk that allocates across both digital assets and the traditional instruments that increasingly wrap them. In that time I have written, or refused to write, something on the order of four thousand internal notes. The composition of those notes has shifted in a way that should alarm anyone who believes that information quality in this market is improving.
In 2017, the modal research note I produced was a valuation document. It had a cash flow model, even if the cash flow was fictional. It had a supply schedule. It had a team section with real names. The analysis was frequently wrong, but the inputs existed. The failure mode was bad reasoning applied to available data.
In 2026, the modal note I am asked to produce is a completeness audit. It asks, in twenty-two fields, what the issuer has disclosed, and the honest answer is usually that the issuer has disclosed four of them, and that three of those four are marketing restatements of the fourth. The failure mode has inverted. The reasoning is frequently sound. The inputs do not exist.
That inversion is the single most important structural change in crypto research this decade, and almost nobody has repriced it.
Why does it matter? Because the analytical frameworks we import from equity and credit research were built for a disclosure regime that is legally mandatory, bureaucratically enforced, and civilly actionable. In the United States, a public company that omits a material related-party transaction from its 10-K exposes itself to securities litigation, delisting, and in extreme cases criminal referral. In the European Union, the Market Abuse Regulation and the Transparency Directive impose a similar architecture. The blank field, in legacy markets, is an anomaly. It is evidence of fraud until proven otherwise.
In crypto, the blank field is the default. It is not an anomaly. It is the baseline state of the universe, and the frameworks that treat it as an anomaly systematically misfire, because they implicitly assign zero informational content to absence when absence is, in this market, the single most informative signal available.
II. The Market for Lemons, Repriced
In 1970, George Akerlof published a paper about used cars that won him a Nobel Prize three decades later. The argument was simple and devastating. When sellers know more about the quality of a good than buyers do, and when buyers cannot verify quality before purchase, the market unravels. Buyers rationally discount every car to the average expected quality. Sellers of good cars, who know their cars are good, refuse to sell at the average price. They exit. The remaining pool degrades. Buyers discount further. The process iterates until the market clears at the quality of the worst remaining seller, or collapses entirely.
Crypto is Akerlof's lot, running in production, at scale, with leverage.
But there is a crucial difference, and it is the difference that makes crypto research interesting rather than merely depressing. In Akerlof's model, the asymmetry is exogenous. The used-car seller did not choose to know more than the buyer; the information asymmetry is a fact of the physical world. In crypto, the asymmetry is manufactured. The issuer chooses what to disclose. Every blank field is a decision made by a specific person, at a specific time, with a specific expectation about what the blank will conceal and what it will cost.
A blank field is not the absence of information. It is information about the intentions of the entity that produced the blank.
This is the pivot on which the entire discipline turns, and it is the pivot that most analysts, human or machine, fail to make. A blank supply schedule is not a neutral observation that we do not know the supply schedule. It is a positive claim: the issuer has determined that disclosing the supply schedule would reduce the valuation more than withholding it would. That is a testable hypothesis. And it has a predictable consequence โ a predictable distribution of future unlocks, a predictable pattern of insider selling behavior, a predictable shape to the price path.
I want to be concrete about the magnitude, because abstractions are how analysts hide from their own conclusions. Between 2021 and 2024, I maintained an internal dataset of token launches that our desk either reviewed, allocated to, or explicitly declined. The sample is not random โ it is biased toward deals that reached a Zurich institutional desk, which is to say it skews toward larger raises and more credible counterparties. That bias makes the finding, if anything, conservative.
The dataset contained 340 launches. For each, I scored disclosure across eleven fields: supply schedule, vesting schedule, cliff structure, treasury composition, auditor identity, contract upgrade authority, jurisdiction, beneficial ownership, revenue disclosure, custodian identity, and governance concentration at genesis.
Projects that disclosed seven or more of those eleven fields at the point of listing had a twelve-month survival rate โ defined as the token retaining a functioning market and the protocol retaining an active development repository โ roughly 2.3 times that of projects that disclosed three or fewer. The effect persisted after controlling, crudely, for raise size and sector, though the sample is too small and too survivorship-biased for the coefficient to be more than directional.
I present it not as a proven law but as what it is: a directional prior built from a biased sample, which is more than most desks have, and considerably less than most desks claim.
The deeper point is the mechanism. Disclosure does not cause survival. Disclosure and survival are both downstream of a third variable โ the presence of an entity with something to lose. A team that discloses its vesting cliff is a team that expects to be held to it. A team that audits its contracts is a team that expects the contracts to be read. A team that names its jurisdiction is a team that expects to be sued in it. The blank field is not a knowledge gap. It is a governance gap wearing a knowledge gap's clothing.
III. The Framework Trap
Here I need to describe a failure mode I have both committed and audited, because it is the mechanism by which the blank-field problem converts itself into institutional losses.
Last year a counterparty sent me a research packet on a mid-cap DeFi protocol. The packet was forty-one pages. It was organized across nine analytical dimensions: technical, token economics, market, ecosystem position, regulatory compliance, team and governance, risk, narrative, and supply-chain transmission. Each dimension had sub-criteria. Each sub-criterion had a rating scale. The document concluded with a composite score and a recommendation.
I ran the document through a simple extraction. Of the roughly one hundred and ten discrete fields the framework defined, ninety-four were marked as not available, not applicable, or not disclosed. The composite score, despite this, was populated to two decimal places. The recommendation was constructive.
This is the framework trap, and it is worth being precise about why it happens, because it is not a failure of intelligence. It is a failure of a specific and predictable kind.
A sufficiently detailed analytical framework will always produce a populated output, regardless of the quality of its inputs, because the framework's completeness becomes a substitute for the analysis's completeness. The human reader, and more importantly the institutional reader, who is often reading under time pressure and looking for a defensible reason to approve an allocation, sees structure and reads it as rigor. Nine dimensions. Forty-one pages. Two decimal places. The document has the shape of analysis. It passes the shape test.
The shape test is the dominant screening mechanism in institutional crypto allocation right now, and it is being exploited deliberately. I have seen decks designed explicitly for it โ the nine-section structure, the obligatory risk section that lists liquidity risk and smart contract risk and regulatory risk in the same font and the same length, the conclusion that acknowledges uncertainty while still recommending a position. This is not analysis. It is set dressing, and it is purchased at a known price from a known set of vendors.
There is a distinction that the industry has collapsed and that needs to be restored, because the two are not the same and the market cannot price them identically.
A null result is information. It is what you produce after you have looked: we tested whether this protocol's fee revenue covers its emissions, we found that it does not, here are the numbers. A null result constrains the space of possible worlds. It tells you something true about the thing you examined.
A null report is the absence of a test. It is a document with empty fields and a populated conclusion. It constrains nothing. It tells you only that someone wrote a document.
The two are structurally indistinguishable to a reader who is scanning for shape rather than content. And because the null report is roughly two orders of magnitude cheaper to produce โ a competent generative pipeline will emit forty pages of it in ninety seconds โ the market has filled with null reports, and the price of the two has collapsed toward each other.
In 2017, at the height of the ICO mania, I ran a stochastic cash-flow model on a token sale that I had been pressured, by the media arm of my own firm, to endorse. The model said the burn rate was mathematically incompatible with a six-month liquidity window under every distribution of cash inflows I could reasonably specify. I declined to sign the endorsement, citing what I called systemic logical failures in the revenue projections, and I passed the technical critique to a small forum where it could circulate. Six months later the founders were indicted. I have told this story before, and I tell it again here for a specific reason: the reason I could run that model in 2017 was that the token sale had published a supply schedule and a treasury address. Two of the eleven fields. That was enough to do arithmetic.
The 2026 equivalent of that deal would publish neither, and the null report written about it would be indistinguishable, in shape, from a rigorous analysis. The industry's analytical infrastructure has improved while its analytical inputs have decayed, and the framework has absorbed the decay without registering it.
IV. What the Blank Field Obscures: A Taxonomy
Before the model, I need a taxonomy, because not all blanks are the same and conflating them is the second most common error I see.
Type I โ Withheld. The data exists. A specific party has it. That party has decided, for reasons we can attempt to model, not to release it. Unlock schedules, cap table composition, and custodian identity are almost always Type I. This is the most informative category, because the withholding decision is a rational act with a rational motive, and the motive is usually visible from the outside.
Type II โ Uncollected. The data exists in principle but nobody has assembled it. Miner revenue decomposition, real order-book depth at a specified distance from mid, stablecoin float by issuer and chain, and cross-venue funding basis dispersion are all Type II. This is not a disclosure failure. It is a research failure. The data is public, free, and on-chain or in exchange APIs, and it sits unread because reading it requires a pipeline most desks have not built.
Type III โ Unknowable. The data does not exist and cannot exist at the time of analysis. The future price of a governance token in a regime where its only value accrual is a fee switch that has not been activated. The behavior of a liquidation engine under a price path that has not occurred. Type III blanks are irreducible, and the correct response to them is not disclosure demands but explicit uncertainty budgeting.
The reason this taxonomy matters is that the three types have opposite implications for position sizing, and the frameworks treat them identically.
A Type III blank is a reason to size down and widen your stop. A Type II blank is a reason to build a pipeline, because you are being paid for work your competitors are not doing. A Type I blank is a reason to walk away, because a Type I blank is a statement by an informed party about the thing you are about to buy.
Most retail-facing research conflates Type I and Type III. It treats withheld unlock schedules as though they were genuinely unknowable, which converts a strategic decision by an insider into an act of God, and it assigns to the resulting uncertainty the sympathetic treatment we reserve for irreducible risk. The unlock schedule is not unknown. It is known, by someone, and that someone's willingness to conceal it from you is the single most reliable negative signal in this asset class.
V. Liquidity Is the Pulse; Policy Is the Brain
Now to the substance of what the blanks conceal. Because the most important thing about a withheld field is not the field itself, it is what the field would have told you about leverage, and leverage is where the cycle is decided.
Liquidity is the pulse; policy is the brain. The pulse can be measured. The brain can only be inferred.
The measurable pulse, in March 2026, consists of a small number of series that any serious desk should be tracking at daily frequency, and that almost no token research document mentions at all.
First, order-book depth at two percent from mid, aggregated across the top three venues by genuine volume after excluding wash flow. This number has been structurally declining in the mid-cap tier for eighteen months. The decline is not a sentiment phenomenon. It is a market-maker inventory phenomenon: post-ETF, the marginal market maker's capital is allocated to instruments with listed options, because options provide the inventory hedging that unlisted altcoins cannot. The spread in the mid-cap tier has widened accordingly. A position that looked like two percent of daily volume in 2023 is now, functionally, four percent of available depth, and the round-trip cost of exiting it has roughly doubled.
Second, perpetual funding basis dispersion across venues. In a healthy market, the funding rate on a given asset converges across venues within a few basis points per day, because arbitrageurs will drill a spread. Persistent dispersion is a liquidity signal: it means the arbitrage capital is not there. I track the cross-venue standard deviation of daily funding as a proxy for the depth of the arbitrage layer, and in the current cycle it has been consistently elevated relative to 2021, which is counterintuitive to anyone who assumed that the arrival of institutional capital would tighten everything. It tightened the top of the book and hollowed out the middle.
Third, stablecoin float as the marginal collateral base. The dollar-denominated stablecoin supply is the settlement layer's working capital. Its growth rate is a better leading indicator of risk appetite in the crypto-native complex than any price series, because it measures the quantity of unencumbered collateral available to take risk, rather than the price at which risk has already been taken. When float growth stalls while prices rise, the rise is leverage-financed, and leverage-financed rises have a specific and predictable resolution.
Fourth, the composition of the ETF creation and redemption baskets, not the net flow number. The net flow is a headline. The composition is a signal. If the marginal net creation is concentrated in a small number of authorized participants, the structure is more fragile than the flow suggests, because the redemption side of the mechanism depends on those participants' willingness to make a market in size during stress. Post-2024, the spot complex has become the price-discovery venue for bitcoin, which means the fragility has migrated from offshore perps to a handful of regulated balance sheets, which is a change in the location of risk rather than a reduction in it.
None of those four series appears in a typical token research document. Not one. The document discusses the roadmap, the team, the partnership announcements, and the token's "utility." That is not research; it is a matching exercise between a template and a marketing site.
I learned this the hard way and then learned it well. In the summer of 2020, during the first wave of yield farming, I built an internal metric I called the DeFi Liquidity Multiplier. The construction was not exotic. I measured, across the major lending and automated market maker venues, the ratio of gross protocol-level claims to net underlying collateral, and I tracked how that ratio interacted with impermanent-loss hedging strategies that were themselves being financed with borrowed assets. The result was a synthetic leverage layer that no single protocol disclosed and no single dashboard displayed. My model said that a thirty percent decline in the price of ETH would produce a cascade that would not be visible in any protocol's own risk report. In June of 2020 the correction arrived and the cascade behaved approximately as the model specified, and I have treated protocol-level liquidity accounting as the primary analytical object ever since.
The relevance to the blank field is direct. The four pulse series are all Type II blanks โ uncollected, not withheld. The metrics that determine whether a position survives are public and unread, while the metrics that determine whether a position gets approved are private and unread. That is a structural misallocation of analytical attention, and it is where I would locate most of the desk-level losses I have reviewed in the past two years.
VI. The Post-Halving Revenue Blank
There is one domain where the blanks are Type II and the consequences are systemic rather than idiosyncratic, and it deserves its own treatment because it is the clearest example of a blank that the entire industry reads as an answer.
Bitcoin's fourth halving cut the block subsidy to 3.125 BTC. This is public. It is the most widely known fact in the asset class. What is far less widely modeled is what it does to the security budget's composition, and the composition is the thing that determines whether the network's security assumptions survive the next two cycles.
Miner revenue has two components: the subsidy and the transaction fees. Outside of congestion events, fees have historically contributed a single-digit percentage of total miner revenue. Post-halving, the subsidy's share of the security budget declines monotonically in BTC terms, while the fiat-denominated security budget depends on the price. This is not a controversial observation. It is arithmetic. But the research documents largely do not do the arithmetic, because the arithmetic produces an uncomfortable output.
Run the series forward under a range of price scenarios, and a threshold emerges. Below a certain price path, the fiat-denominated security budget declines faster than the network's hashrate, which means the marginal miner's revenue per unit of hash โ the hashprice โ compresses below the marginal cost of the least efficient installed machine. That machine shuts off. The hashrate falls. The difficulty adjusts. The network re-equilibrates at a lower security level. This is how the mechanism is supposed to work, and it does work. The blank field is what happens to the distribution of that hashrate as it re-equilibrates.
Mining is an industrial business with industrial economics, which means it is subject to the same consolidation dynamics as any other capital-intensive commodity extraction industry. When hashprice compresses, the operators who survive are the ones with the lowest energy cost and the deepest balance sheet. Those operators participate in pools, and the pool landscape has been concentrating for years. The concentration is measurable, and I have measured it: the share of blocks attributable to the small number of largest pools has been trending upward through the halving transitions, with a step change around each subsidy reduction.
The decentralized-consensus claim for Bitcoin rests on a specific structural assumption: that no coalition of miners and pools can coordinate to censor or reorganize. That assumption is not a constant. It is a function of the distribution of hashpower, which is a function of the cost structure, which is a function of the subsidy and the price. If the distribution continues to concentrate, the assumption degrades โ not suddenly, not visibly, and not in a way that produces a headline, but continuously and in one direction.
I have held this view since the halving and I will state it plainly, not as a prediction but as a structural read: the post-halving revenue compression is converting Bitcoin's decentralization from a technical property into a consensus narrative, and the blank field is the absence of anyone modeling the transition. The data exists. Pool-level block attribution is public and timestamped. The modeling simply has not been done by the parties with an incentive to publish it, and the parties with the capability to publish it have an incentive not to.
That is a Type II blank that has been allowed to become functionally a Type I blank, not through concealment but through collective inattention. It is the most expensive kind.
VII. The Compliance Blank: What MiCA Actually Filters
The European Union's Markets in Crypto-Assets regulation was marketed, accurately in part, as the arrival of clarity. It establishes a licensing regime for crypto-asset service providers, imposes reserve and custody requirements on stablecoin issuers, and creates a supervisory architecture that, unlike the previous patchwork, has a single rulebook and a defined enforcement mechanism.
Clarity has a price, and the price is disclosed in a way that most market commentary does not read.
The stablecoin provisions of MiCA require issuers of asset-referenced and e-money tokens to hold reserves backing the outstanding supply, with constraints on the composition of those reserves, including limits on the proportion held in bank deposits relative to high-quality liquid assets. The requirements are not unreasonable in themselves. They are, however, expensive to satisfy continuously and to evidence to a supervisor continuously, and the cost is largely fixed rather than proportional. A stablecoin with a hundred million in circulation and a stablecoin with ten billion in circulation face broadly similar compliance architectures โ the same legal opinions, the same audit function, the same reporting infrastructure, the same supervisory engagement.
For the licensing regime, the pattern is the same. Authorization as a crypto-asset service provider requires legal counsel in at least one member state, an audit function, governance documentation, capital adequacy under the applicable prudential framework, and an ongoing supervisory relationship. Even for a lean applicant, the aggregate cost of the authorization process and the first year of compliance runs well into six figures in euros, and for applicants with custody or exchange functions it runs higher. Ongoing compliance is a recurring cost with a headcount attached.
This produces a filtering effect that is mechanical rather than conspiratorial. A project with a hundred million in treasury can absorb the compliance cost without changing its roadmap. A project with five million in treasury cannot. The EU regime therefore functions, whatever its stated intent, as a minimum balance sheet requirement for market participation in the largest regulated jurisdiction in the world.
The blank field enters here in a specific way. Small projects do not disclose their MiCA strategy because they do not have one, and the absence of a disclosed compliance path is, in the current regime, a near-certain indicator that the project has concluded, correctly, that it cannot afford one. This is a Type I blank with a positive informational content that is consistently misread as Type III. Analysts look at the empty regulatory section and conclude that the regulatory situation is uncertain. It is not uncertain. It is determined, and the determination is unfavorable.
I would put the structural consequence this way: MiCA does not eliminate bad actors from the European market. It eliminates small ones, and it does so with a threshold that is legible to anyone willing to estimate the fixed cost of authorization. The clarity is real. So is the consolidation. The two are the same mechanism viewed from different ends.
VIII. The Identity Blank: Why the Concept Refused to Ship
There is a third blank worth examining, because it is the cleanest demonstration that some blanks persist not because the technology is missing but because filling them is individually irrational.
Soulbound tokens were proposed as the primitive that would bring reputation, credentials, and eventually credit to public blockchains. The argument was straightforward. If a token is non-transferable and permanently bound to an address, it can function as an attestation: a verifiable, portable record that a specific address has done a specific thing. Layer enough attestations and you have a portable identity, and with portable identity you have the substrate for undercollateralized lending, which is the thing that DeFi has never solved and that legacy finance solves with credit bureaus.
The concept has now been circulating for roughly three years. The realized deployment is negligible relative to the conceptual volume. And the reason is not cryptographic โ the cryptography works, and the zero-knowledge primitives that would allow selective disclosure of attestations without revealing the full record have matured considerably, to the point where the privacy objection is technically answerable.
The reason is that the attestation market has a specific equilibrium structure, and that equilibrium is empty.
Consider what an attestation is worth to the party holding it. A positive attestation โ this address has repaid nine loans on time โ is valuable to the holder, because it unlocks credit. A negative attestation โ this address has defaulted โ is not valuable to the holder, because it only closes doors. The rational holder therefore discloses positives and withholds negatives. The lender, knowing this, and knowing that the holder controls disclosure, discounts every disclosed positive by the base rate of withheld negatives. The discount reduces the value of the positive attestations, which reduces the incentive to accumulate and disclose them, which reduces the population of attestations, which raises the discount. The equilibrium is degenerate. Nobody discloses enough for the market to function, and because nobody discloses, there is no reason to be the first to disclose.
This is Akerlof again, as it always is. And it is why the permanent, public, non-transferable record is a concept rather than a product. The technology of soulbound identity is not the obstacle. The obstacle is that a credibly permanent record of one's worst moments is a liability whose value is strictly negative to the only party who controls its creation. The blank field persists because filling it is individually dominated, and no amount of protocol design changes the individual incentive.
The narrow exceptions are instructive. Attestations spread where the disclosing party is not the individual: employment credentials issued by an employer, academic credentials issued by an institution, sanctions screening issued by a compliance vendor. Those work because the issuer, not the subject, bears the reputational cost of a false attestation. That is a small market. It is a real market. It is not the substrate for undercollateralized lending, and the industry's continued implication that it will become one is a category error that has survived three years of contrary evidence.
IX. Sizing a Position Against Nothing
I want to close the core analysis with something operational, because a diagnosis of information decay is only useful if it changes what you do with capital. I am going to sketch the model I use, and I am going to be explicit that the numbers are illustrative, because presenting illustrative numbers as calibrated is precisely the sin I have been describing for four thousand words.
The classical sizing rule is Kelly: allocate the fraction of capital that maximizes the expected logarithmic growth rate, which for a binary bet with payoff odds b, win probability p, and loss probability q = 1 - p, is f* = (bp - q) / b.
Suppose an analyst estimates a win probability of 0.56 and a payoff ratio of 1.5:1. Then f* = (1.5 ร 0.56 - 0.44) / 1.5 = 0.267, or roughly 27 percent of capital. That is already an aggressive number, and any practitioner would scale it down. The scaling usually applied is fractional Kelly โ a half or a third โ to compensate for the fact that p is an estimate rather than a known parameter.
But fractional Kelly addresses parameter variance. It does not address parameter bias. The systematic error in crypto research is not that our win probabilities are noisy around the truth. It is that they are computed from disclosed data, and the disclosed data is a non-random sample. This is the missing-data problem, and it has a specific structure that statisticians have a name for.
Data missing completely at random is benign: the missingness is uncorrelated with the value, so the observed sample is an unbiased, merely smaller, version of the population. Data missing at random is handleable: the missingness depends on observable covariates, and the appropriate reweighting recovers the parameter. Data missing not at random โ the value of the missing observation depends on the fact that it is missing โ is the pathological case, and it is the case that describes every blank field in crypto.
Supply schedules are withheld precisely by the issuers whose schedules are the worst. Audits are missing precisely where the contracts are weakest. Custodian identities are concealed precisely where custody is worst. The missingness and the value are the same variable. You cannot correct for MNAR by shrinking your position size, because the bias is in the sign of the edge, not its magnitude.
So I apply a disclosure ratio, d, defined as the share of the analytically decisive fields actually disclosed, and I use it to shrink the edge itself rather than the size. Formally, if the raw edge is the difference between the estimated win probability and the coin-flip, then the effective win probability is p' = 0.5 + d(p - 0.5). With p = 0.56 and d = 0.3, we get p' = 0.518 and q' = 0.482, and the Kelly fraction becomes f* = (1.5 ร 0.518 - 0.482) / 1.5 = 0.197, or roughly 20 percent โ a reduction of about a quarter.
That number understates the correction for two reasons, and the understatement is the point. First, d should not be computed across all fields. It should be computed across the fields that are analytically decisive for the specific thesis. An unlock schedule is decisive for a newly listed token and irrelevant for a protocol with a fully circulating supply. A weighted disclosure ratio, where the weights are the sensitivity of the valuation to each field, is smaller than the unweighted one, sometimes dramatically.
Second, and more importantly, the disclosure ratio should not be applied as a linear shrink at all when the missingness is concentrated in fields that determine solvency rather than profitability. If the withheld field is the treasury runway, the appropriate posterior is not a smaller positive edge. It is a substantially enlarged probability of a zero. I model that as a separate ruin probability, subtracted from the terminal payoff rather than blended into the win rate, and the effect on the optimal fraction is discontinuous. Positions that survive a shrunken edge do not always survive an added ruin term, because Kelly is exquisitely sensitive near the boundary where expected log growth crosses zero.
In practice the rule I use is blunt, and I have not found a reason to soften it: when the decisive fields are Type I blanks and the thesis depends on them, the correct position size is zero. Not small. Zero. Because the blank is a forecast that someone with better information has made about the answer, and the forecast points the same direction every time.
X. The Contrarian Case: Incompleteness Is the Pricing Mechanism
Here I need to argue against my own thesis for a moment, because the position I have built โ that blanks are informative and should be penalized โ has a complacent version that is also wrong, and the complacent version is where the marginal analyst stops.
The complacent version goes like this: disclosure is good, more disclosure is better, the market is inefficient because disclosure is inadequate, and therefore the analyst who demands disclosure is doing God's work and will be rewarded by an eventual repricing toward a complete-information equilibrium.
That is a legacy-finance import, and it is wrong on both the mechanism and the magnitude.
On the mechanism: mandatory disclosure in equities works because there is a state with a monopoly on violence that can compel it and punish its absence. Crypto has no such state at the protocol layer. The relevant enforcement is jurisdictional and partial, and the entities that most need disclosure are precisely the ones that will route around any jurisdiction that imposes it. The equilibrium under any voluntary or partially-enforced disclosure regime is not convergence toward completeness. It is a segmented market: regulated venues with full disclosure and thin liquidity, and unregulated venues with no disclosure and most of the volume. That segmentation already exists, and MiCA has accelerated it rather than reversed it.
On the magnitude: the reason the industry has not converged on disclosure is not that nobody has demanded it. It is that the demand has been made continuously since 2017 and the supply has not materialized, which means the demand is not being met by the mechanism that the complacent view assumes. The persistence of the blank field across three cycles, four regulatory regimes, and one institutional ETF complex should be read as evidence that the blank field is load-bearing โ that it is part of how the market clears, not a friction that the market has failed to remove.
If that is right, the implications invert the standard view.
First, the information risk premium โ the discount the market applies to an asset whose disclosure is poor โ is not a free lunch for the analyst who is willing to do the work. It is compensation for a risk that cannot be diversified away, and in a portfolio context the correlation of that risk across positions is high, because non-disclosing issuers share a common factor. A book of twelve under-disclosed positions is not twelve diversified bets. It is one bet, twelve times.
Second, mandatory disclosure would not primarily help the sophisticated analyst. It would primarily help the passive allocator, by making the market legible at scale, which would reduce the dispersion of returns and therefore reduce the value of research. The sophisticated analyst has a material interest in the blank field that he is rarely honest about. I include myself in this. A market that is legible to everyone is a market in which edge is scarce.
Third, and most counterintuitive: the true value of the blank field is not that it hides bad assets. It is that it hides the fact that the entire asset class is priced off a consensus rather than an accounting. Value is a consensus, not a fundamental truth. The blank fields are the places where the consensus is thinnest, and the price action around them is where the consensus gets renegotiated. If the blanks were filled in, the renegotiation would still happen, but it would happen in the accounting, at a lower amplitude, over a longer time, and with less opportunity for anyone paying attention.
The blank field is not a defect in the market. It is the market's resolution mechanism, operating in the open, one undisclosed unlock at a time.
I do not like this conclusion. I have spent two decades arguing against it. But the evidence of three cycles is that the blanks survive because the participants with power prefer them to survive, and the analyst who builds a career on demanding their removal is building on a demand the market has declined, in public, every quarter for nine years.
XI. The AI Research Flood and the Repricing of Information Gain
There is one force operating on this system in 2026 that did not exist in 2017, and it changes the arithmetic of the blank field decisively.
Until roughly 2023, the production of a plausible research document required human labor, and human labor is priced. A forty-page null report cost an analyst two weeks, or a junior analyst three, and that cost imposed a floor under the supply of documents and therefore a floor under their value. The floor was low, but it was a floor.
That floor is gone. A generative pipeline with access to a project's public materials and a template will produce a forty-page document with nine analytical dimensions, a risk matrix, and a populated composite score in under a minute, at a marginal cost near zero. I know this because I have built the pipeline internally, deliberately, for the purpose of stress-testing our own output. The documents it produces are structurally indistinguishable from a substantial fraction of what is sold as sell-side crypto research in 2026. I am not being rhetorical. I tested it against a sample of externally produced reports, stripped the attributions, and showed both to three colleagues with a combined forty years of experience. Their identification accuracy was not statistically distinguishable from chance on the subset of reports that contained no primary data.
What follows from this is not that AI research is bad. It is that the price of a document has decoupled from its information content, and the two are now priced almost entirely separately.
Information content, in the sense that the search and publishing ecosystem has codified, is the quantity of claims in a document that the reader did not already know and cannot trivially verify. The 2026 search quality frameworks call this information gain, and the reason they call it that is that a document with zero information gain is a document that adds nothing to the corpus โ it is a restatement of the corpus in different words.
The null report has zero information gain by construction. And it is now free to produce. So the industry is producing an unbounded quantity of it, and the market is doing what markets do when the supply of a zero-value good becomes infinite: it is repricing the entire category toward zero.
The consequences for the analyst are structural, and I have been repositioning for them for two years.
In 2024, after the spot ETF complex opened, I shifted the majority of my analytical time from token-level work to liquidity-level work, and I did so for a reason that I can now state cleanly. Token-level analysis is precisely the domain where the marginal document is now generated at zero cost, which means the marginal document's value is zero, which means the return to producing it is zero. Liquidity-level analysis is the domain where the inputs are Type II blanks โ uncollected, public, and requiring a pipeline โ which means the marginal document requires capital expenditure to produce, which means the supply is constrained, which means the return is not.
I spent the back half of 2024 and the whole of 2025 backtesting a specific hypothesis with a Swiss quantitative fund: that the integration of algorithmic execution with crypto liquidity venues would compress retail-accessible arbitrage opportunities by roughly forty percent by 2026. The backtest confirmed the direction and roughly the magnitude. The mechanism is not mysterious. Algorithmic market makers quote tighter and reprice faster than discretionary traders, so the windows that a human could exploit are closed before the human sees them. What the backtest also showed, and what I had not fully anticipated, is that the compression is uneven across signal types. Price-based arbitrage compressed substantially. Narrative-based arbitrage โ the exploitation of a gap between what a document claims and what the data shows โ compressed far less, because the documents are the input and the documents are now generated carelessly at industrial scale.
That is where the residual edge lives, and it has a specific shape. The last exploitable inefficiency in crypto is the gap between what a generated document asserts and what the public data actually shows, and the blank field is the widest part of that gap.
XII. What to Watch
I do not want to end with a summary, because a summary would be exactly the kind of zero-information-gain artifact I have spent this article dissecting. Instead, here are the five series I am tracking, in order of what would change my positioning if they moved.
First, the emergence of research provenance attestations. If a market develops for on-chain, verifiable attestations of a research document's inputs โ a signed record of which data series were used, at what timestamps, from which sources โ then the null report becomes detectable at the point of consumption, and the price of the two categories separates. This is a Type II blank in the research layer itself, and I would expect the first serious attempt to come from a venue that needs to defend its own research in a dispute, not from a research house.
Second, disclosure indices as a rating input. If a major index provider or a regulated fund administrator begins to incorporate a disclosure score into eligibility criteria, the blank field acquires a price at the index level, and the marginal issuer faces a genuine cost for concealing. That is the only mechanism I can see that would move the equilibrium, and it requires the buyers of crypto exposure, not the sellers, to demand it.
Third, the fee share of miner revenue. This is the single most informative public series in the asset class and it is almost never discussed. If the fee share does not structurally rise over the next two halving intervals, the security budget's composition becomes a live constraint rather than a theoretical one, and the concentration dynamics I described earlier accelerate on a schedule that is known to everyone and modeled by almost no one.
Fourth, the count of authorized crypto-asset service providers in the European Union. The number is public and it is a direct measurement of the compliance filter's tightness. If it plateaus below the level implied by the pre-MiCA population of operating entities, the consolidation thesis is confirmed at the jurisdiction level, and the surviving entities' market power is worth pricing.
Fifth, the first genuine negative-attestation use case. The soulbound identity market will not be built on the voluntary disclosure of positive credentials, and I would watch for the first product where the issuer of the attestation, not the subject, has an economic stake in its accuracy โ a collateral registry, a custody attestation, an audit attestation. That is the only configuration where the degenerate equilibrium I described can be broken, because it moves the cost of a false attestation from the party with the least incentive to bear it to the party with the most.
And the question I would put to anyone still holding a position justified by a document with a populated composite score and ninety-four empty fields is this. If every disclosed field is chosen by the seller, and every withheld field is chosen by the seller, and the seller knows which is which โ then what, precisely, is the thing you believe you are pricing? The market has a model for a lie. It even has a model for a rumor. It has never had a model for a blank, and the blanks are the only thing this cycle has been selling.
