Hook
The data indicates a failure mode, not a data gap.
On a Tuesday, a nine-dimension analytical framework was executed against a crypto asset. Forty-seven discrete cells were produced across technical, tokenomic, market, ecosystem, regulatory, team, risk, narrative, and supply-chain dimensions. Every cell returned the same value: N/A โ insufficient information. No title. No source. No information points. No project identified. No timestamp.
The correct response to that output is not to fill it in. It is to publish it.
I have run audit pipelines for nine years across three market cycles. The single most expensive defect I have ever found was not a reentrancy bug, not an oracle manipulation, not a mispriced liquidity pool. It was a report that looked complete. In 2017 I sat in a Sydney conference room and watched a counterparty sign a term sheet on the strength of a forty-page deck in which thirty-one pages were financial projections generated from a spreadsheet with no source column. The tokenomics were internally consistent. The arithmetic was correct. The inputs were invented. The project lost 94% of its value in eleven weeks.
The document in front of me โ the null report โ is the first artifact of its kind I have seen produced voluntarily. It is a machine-readable confession. It says: I was asked nine questions, I had no evidence, and I declined to answer. In a market where the median research note is 60% adjectives, that refusal is worth more than the note.
So I am going to treat the null report as a specimen. Not as an error to be corrected, but as evidence of a pathology that has been quietly compounding in crypto research since 2020, and which the current sideways tape has finally made visible.
Context
Crypto research is an industry with a cost structure and a revenue model. Understanding the second one explains almost everything about the first.
Between 2017 and 2021, sell-side crypto research was funded by token issuers, exchanges, and venture funds who needed narrative support for positions they already held. The output was long, expensive, and directionally pre-committed. Between 2021 and 2024, that model collapsed under its own weight: too many reports, too few readers, and a terminal credibility problem after the Terra and FTX cycles demonstrated that the most effusive coverage clustered precisely around the assets that failed hardest.
What replaced it was not better research. It was cheaper research. Large language models reduced the marginal cost of a 4,000-word protocol report from roughly forty analyst-hours to roughly four minutes. By 2025, the volume of published crypto analysis had increased by an order of magnitude while the volume of verifiable claims inside it had, by my sampling, fallen.
This is the part most readers miss. The problem is not that AI writes badly. The problem is that AI writes plausibly, and plausibility is the exact failure mode that crypto research cannot tolerate. A human analyst who does not know a project's unlock schedule will write "details unclear." A language model that does not know the unlock schedule will write "the team's allocation is subject to a standard vesting schedule, aligning long-term incentives with holders." Both sentences are generated from the same absence of data. Only one of them is honest.
The current market regime accelerates the pathology. In a trending market, narrative does the work: a token goes up, and the report that predicted it going up gets archived as vindication. In a sideways market โ which is what we have had for the better part of this cycle โ narrative stops paying. Capital stops rotating on story and starts rotating on measurable things: emission schedules, real fee revenue, unlock cliffs, sequencer margin, collateral quality. Readers no longer want to know what a protocol says it is. They want to know what its balance sheet is.
That shift is why the null report matters. A framework built for a narrative market rewards confident prose. A framework built for a consolidation market rewards nothing but evidence. The nine dimensions โ technical, tokenomic, market, ecosystem, regulatory, team, risk, narrative, supply-chain โ are not nine opinions. They are nine evidentiary requirements. Each one has a minimum input. Without that input, the dimension does not produce a weak answer. It produces no answer.
The analyst who returned the null report understood something that most of the industry still does not: a framework with empty cells is a diagnostic instrument, not a broken one. It tells you exactly where the evidence stops. That is the single most useful piece of information in any investment process, and it is the one piece that almost nobody publishes, because it is the one piece that cannot be sold.
I want to be precise about the standard I am applying here, because it is the standard I was trained on. My MS is in financial engineering. My first professional engagement, in late 2017, was a six-week tokenomics audit of a project calling itself Ethereum Classic Network, promising 1,000% APY. I modelled their liquidity pools against SEC securities law, and I found that 40% of the token supply was unvested and unscheduled โ no cliff, no lockup, no disclosure. That single line item, in isolation, was enough to terminate the engagement. The project was delisted from local exchanges within the quarter.
I did not need the other eight dimensions. I needed one that had a verifiable input. That is the discipline the null report enforces: you do not get to average evidence across dimensions. One empty cell in a critical dimension invalidates the conclusion, not the framework.
Core
1. The anatomy of a fabricated cell
Let me show what the alternative to the null report looks like, in the exact format the framework produces. Here is a token supply table as it typically appears in published research, and beside it the same table as it appears in a null report.
| Category | Published Research | Null Report | |---|---|---| | Team | 18% โ 4-year vest, 1-year cliff | N/A โ no disclosure | | Early investors | 22% โ 2-year vest | N/A โ no disclosure | | Community / liquidity | 40% โ via emissions | N/A โ no disclosure | | Treasury / ecosystem | 20% โ DAO-controlled | N/A โ no disclosure |
The left column is not a lie in the ordinary sense. It is almost certainly a paraphrase of a whitepaper. The problem is that a whitepaper allocation table is a claim, not a state. The state lives in the token contract, in the vesting contract, in the multisig, and in the block explorer. The published column has copied the claim and presented it as the state. That is the definition of a bug in an analytical system: the output is derived from the wrong source of truth.
I have a rule that has never failed me. If a supply table cannot be reconstructed from four things โ the token contract address, the deployer address, the vesting beneficiary list, and the block height of deployment โ then it is marketing, not data. Not one of those four requires permission to access. They are all on-chain. The reason the published table exists in its current form is not that the on-chain data is unavailable. It is that the on-chain data is worse news.
In the 2017 Ethereum Classic Network case, the on-chain reconstruction took me three days and produced a number the deck did not contain: 40% of supply sat in eleven addresses, nine of which had never transacted, and two of which were the same address under different labels. That is what an unlock schedule looks like before anyone has written a document about it. It looks like silence.
2. The Howey table is not a formality
The regulatory dimension of the framework contains a four-element securities test. In most published research it is completed with a single sentence: "The token is sufficiently decentralized and therefore unlikely to be classified as a security." That sentence has no evidentiary content whatsoever.
Here is the same dimension done properly, with the actual dependencies.
| Howey Element | Evidentiary Requirement | Failure Mode | |---|---|---| | Investment of money | Trace sale proceeds to a receiving address | Proceeds routed through a mixer or unlabelled bridge | | Common enterprise | Demonstrate pooled capital with shared exposure | Token sold as individual utility licence | | Expectation of profit | Show marketing language promising returns | Marketing scrubbed post-launch; archives deleted | | Efforts of others | Identify a core team whose work drives value | Team anonymous and unaudited |
The fourth element is where most projects quietly fail, and where most research quietly stops. If a protocol's value depends on a team shipping a roadmap, then the token is a claim on that team's labour. The only defence is that the team's labour is not material โ which is another way of saying the protocol already works without them. Very few do.
My 2017 audit applied exactly this table. The project's marketing materials promised 1,000% APY in a jurisdiction that had not yet classified yield-bearing tokens. Elements one, two, and three were satisfied on the face of the documents. Element four was satisfied by the existence of a nine-person team whose entire function was to operate the yield mechanism. The conclusion was not ambiguous. It was a security, sold without a prospectus, in a jurisdiction that requires one. That is why the delisting was immediate and uncontested โ the exchange's counsel reached the same conclusion in a single afternoon once the table was on the page.
The Howey test is not a legal opinion. It is an evidence checklist. If you cannot complete all four rows with addresses and documents, you have not completed the dimension.
3. The rounding error that was worth two million dollars
In the summer of 2020 I independently audited the Compound Finance governance contract v1. This is the part of the article where I stop talking about research and start talking about code, because the two are the same problem expressed in different languages.
Interest accrual in Compound-style markets works by maintaining a borrow index that grows monotonically. Each block, the contract computes a borrow rate from utilisation, scales it by the block delta, and adds it to the index. The critical implementation detail is mantissa scaling: rates are stored as 1e18-scaled integers, and every division truncates. Truncation is not a bug in itself. Truncation accumulated across asymmetric operations is a bug.
I rebuilt the accrual loop in Python, block by block, using the same integer semantics rather than floating point. This matters. A float-based replication will show you a smooth curve and hide the defect entirely. You have to model the truncation to see it.
What the integer replication showed was a directional asymmetry: the truncation error in the borrow-side accumulation and the truncation error in the supply-side distribution did not cancel. They compounded in one direction. Under normal volatility the drift was negligible โ fractions of a basis point per epoch. Under high volatility, with utilisation pinned near the kink and block times compressed, the drift became exploitable: an actor with sufficient capital could repeatedly open and close positions at the boundary, harvesting the accumulated truncation as if it were yield.
My model put the extractable value at approximately $2 million over a sustained high-volatility window. I disclosed it to the core developers before any public mention. It was patched.
I want to be clear about what that exercise proved, because it is frequently misread. It did not prove that Compound was insecure. It proved that technical elegance and correctness are orthogonal properties. The contract was elegant. It was audited. It was correct in the sense that every function did what its comment said. And it still leaked value, because the leak was in the gap between two correct functions.
The null report's technical dimension is the equivalent of my Python replication. It asks: is there an audited artifact I can read, and does it match the deployed bytecode? If the answer is no, the dimension returns N/A. Not "likely secure." Not "audited by a reputable firm." N/A. Because a marketing claim about an audit is not an audit, and an audit of commit a3f9c1 is not an audit of the bytecode currently receiving deposits.
4. Terra: the seigniorage mechanism was legible before it failed
In May 2022 I spent three days on LunaScan reconstructing the failure of the algorithmic peg. I am going to describe the mechanism precisely, because the popular account โ "a bank run happened" โ is not an explanation. A bank run is a description of a crowd. The mechanism is a description of a system.
UST maintained its peg through a mint-and-burn arbitrage against LUNA. If UST traded below $1, arbitrageurs could burn $1 of UST for $1 of LUNA, sell the LUNA, and pocket the difference. This works if and only if LUNA has depth. The peg was therefore not collateralised by assets. It was collateralised by the continued willingness of the market to absorb new LUNA supply at a non-declining price.
That is a reflexive structure, and reflexive structures have a specific mathematical signature: the collateral value is a function of the supply, and the supply is a function of the collateral value. Write it out and the derivative is positive. Positive feedback. The system is stable only in the direction of expansion.
When I traced the on-chain flows, the sequence was not mysterious. The bridge outflows preceded the peg break by roughly 36 hours. Liquidity in the Curve pool thinned first. The mint-burn arbitrage then began minting LUNA at an accelerating rate to absorb the UST being sold, which increased LUNA supply, which decreased LUNA price, which increased the UST quantity required to restore the peg, which increased the minting. The loop closed. Approximately $40 billion of value was destroyed in a structure that had no external revenue and no external collateral.
Here is the point I want to make about research, not about Terra. Every input required to reach that conclusion was public before the collapse. The mint-burn function was in the code. The Curve pool balances were on-chain. The absence of an external revenue stream was disclosed. The research industry's failure was not an information failure. It was a dimension-completion failure. The tokenomic dimension was filled in from the whitepaper's description of the mechanism, and the mechanism was described accurately โ but the sustainability sub-dimension, which requires an external revenue line, was left blank and never flagged as blank.
That is the difference between the null report and the Terra research consensus. The null report would have printed: real revenue as a share of yield โ N/A. And that single N/A, sitting in a table next to an advertised 19.5% APY, would have been the most alarming number in the document.
5. MetaCity and the arithmetic of redistribution
In 2023 I evaluated an NFT project called MetaCity, which marketed virtual real estate yields. I did not read the marketing. I asked for the contract and I pulled the holder distribution.
Two findings, both trivial to produce, neither present in any coverage I could find.
First: the yield. I traced every distribution transaction to its funding source. There was no external revenue. Every payout originated from the primary sale contract or the secondary royalty stream. The "yield" was a redistribution of new buyer funds to earlier buyers, net of a protocol fee. That is not a yield model. It is a transfer function with a haircut.
Second: the holders. Using standard clustering heuristics โ common funding sources, transaction graph proximity, gas-price fingerprinting โ approximately 95% of holder wallets resolved to clusters controlled by the deploying team or entities funded by it.
| Metric | Claimed | Measured | |---|---|---| | External revenue | Present | Zero | | Yield source | Asset performance | New buyer inflow | | Independent holders | "Community-owned" | ~5% of wallets | | Team-controlled float | Undisclosed | ~95% of wallets |
Trading volume fell 60% within three weeks of publication. I take no satisfaction in that. I note it because it demonstrates something about the market's information state: the data was always available. It was simply never asked for. The null report's ecosystem dimension asks for DAU/MAU and retention. MetaCity could not produce either, because 95% of its "users" were the same entity. The dimension would have returned N/A, and the N/A would have been the finding.
6. The interest rate models are administered prices wearing a market costume
I have a long-standing technical position that I will now substantiate rather than assert: the interest rate models used by the two largest lending markets are not market mechanisms. They are governance-set step functions.
Consider the canonical utilisation-based model. Utilisation U is defined as borrows divided by borrows plus available liquidity. The borrow rate is then a piecewise linear function of U with a kink: a shallow slope below the kink, a steep slope above it. There are, in the standard implementation, four parameters: base rate, slope one, slope two, and the kink location.
Four constants. Set by a governance vote. Changeable in a single transaction.
Now compare that to a market-cleared rate. In a market-cleared rate, the price is an output of the interaction between borrowers and lenders, and it moves continuously as their preferences change. In the Aave/Compound model, the price is an output of four numbers that a token-holder vote selected, and it moves only when utilisation crosses a threshold or when governance changes a constant.
This is not a criticism of the design. The design is a reasonable engineering compromise for a system that must quote a rate at every block without an order book. But it is a category error to describe the resulting rate as the market price of liquidity. It is an administered price with a feedback loop attached. The feedback loop is real โ high utilisation does push the rate up โ but the mapping from utilisation to rate is a political artifact, not a discovery.
The practical consequence matters for anyone doing risk work. When you model a lending market's tail behaviour, you are not modelling a market. You are modelling a governance process with a four-parameter output, and the tail risk is concentrated in the scenario where utilisation sits just below the kink and the kink is set too high to attract repayment. I have seen this exact configuration in three separate deployments. In each case the model was "calibrated" โ meaning the four constants were chosen to produce an attractive headline APY at the utilisation level observed on the day of launch.
A parameter chosen to produce a desired output is not a parameter. It is a target. And a target embedded in a rate model is a bug that reports itself as a feature.
7. The blob arithmetic, and why the fee floor doubles
The Layer2 dimension of the framework asks for cost structure. Almost nobody completes it, because almost nobody runs the arithmetic on data availability.
Post-EIP-4844, rollups pay for blob space in a separate fee market with its own target and maximum. The target is the level at which the blob base fee stays at its minimum; above target, the base fee rises exponentially; below target, it decays. The design is elegant and it has one property that the marketing literature consistently omits: the blob fee is a function of aggregate demand across all rollups, not of any individual rollup's efficiency.
Run the numbers. A 12-second slot produces 7,200 blocks per day. At the post-Pectra target of six blobs per block, that is 43,200 blobs per day, each carrying 128 KB, for roughly 5.5 GB of daily data availability capacity. Every rollup on every chain that settles to this layer draws from that single pool.
Now apply a demand curve. Rollup DA consumption has grown roughly with the sum of their transaction volumes, and transaction volumes across the major rollups have grown at a rate that has, in every prior capacity expansion I have observed, consumed the new headroom within four to six quarters. When aggregate demand crosses the target, the base fee does not rise linearly. It rises by a fixed multiplicative factor per block while demand exceeds target. A sustained overshoot of two times target does not double the fee. It moves the fee to a level determined by the exponential update rule, which is substantially more than double.
| Aggregate DA Demand vs Target | Blob Base Fee Regime | Rollup Cost Impact | |---|---|---| | Below target | Minimum fee, decaying | Effectively free | | At target | Floor | Negligible | | 1.5ร target | Exponential ascent begins | Measurable per-tx cost | | 2ร target sustained | Fee escalates until demand clears | Doubling or more |
The conclusion I hold is uncomfortable for the sector: the current cheap-fee environment on Layer2 is not an efficiency gain. It is a subsidy paid by unused capacity, and unused capacity is a temporary condition. When the blob market saturates โ and on my modelling that is a two-year horizon from the last capacity expansion, not a decade โ every rollup's cost base resets upward simultaneously, because they are all bidding for the same scarce resource. The rollups that survive that reset will be the ones with fee revenue that exceeds their DA cost at the saturated price. I have looked at the disclosures. Most do not publish that number.
8. The Bitcoin fee market, and the narrative that saved it
I will close the technical section with the position that has cost me the most arguments.
Bitcoin's security model is a fee market attached to a shrinking subsidy. The subsidy halved to 3.125 BTC per block in 2024 and halves again to 1.5625 BTC in 2028. For the security budget to remain constant in dollar terms, fees must absorb the difference. For two decades, the working assumption was that fees would grow into that gap through transaction volume. Transaction volume did not grow into that gap. Block space demand from payments is structurally capped by the fact that Bitcoin is a settlement layer, not a payments network, and the Lightning capacity that was supposed to generate the demand has not produced fee pressure at the base layer.
What produced fee pressure at the base layer was inscriptions. The Ordinals protocol and the tapscript witness-data inscription format created a new class of block-space consumer that pays fees and does not compete with monetary transactions for the same UTXO set. In the periods when inscription activity peaked, fee revenue reached multiples of the subsidy, and miners who had modelled their economics on subsidy-only were, briefly, wrong in a profitable direction.
I am not making an aesthetic argument about JPEGs on Bitcoin. I am making a balance-sheet argument. Remove the inscription demand and the security budget in the post-2028 regime is a number that requires a conversation nobody in the ecosystem is prepared to have. The inscription wave is not decoration. It is, at present, the only demonstrated mechanism by which base-layer fee revenue scales independently of the subsidy. That is a load-bearing fact, and treating it as an embarrassment rather than as infrastructure is the kind of sentiment-driven reasoning that the null report exists to prevent.
9. Why the empty framework is the audit trail
I want to return to the artifact itself, because it contains a lesson that generalises.
A framework with forty-seven cells, all returning N/A, is not a failed analysis. It is a complete record of what could not be established. That record has three properties that a filled-in report does not.
It is falsifiable. Anyone can take the null report, supply the missing inputs, and re-run it. The output will change, and the change will be attributable to specific inputs.
It is composable. A null report from one analyst and a null report from another can be merged, because the missing cells align. Two confident reports from two analysts usually cannot be merged, because their fabricated cells conflict.
It is auditable. The most valuable line in any research document is the line that says what the author did not check. That line tells the reader exactly how much weight the document can bear.
In the absence of data, opinion is just noise. The null report is the only document in the stack that is not noise, because it does not contain an opinion. It contains a boundary.
Contrarian
Here is where I have to be honest about what the bulls got right, because the forensic posture has a failure mode of its own, and it is the failure mode I am most at risk of.
The bearish read on the null report is obvious: a research pipeline that returns nothing is a broken pipeline, and the correct response is to fix the inputs, not to celebrate the emptiness. I have made that argument to clients. It is a good argument. But it is incomplete, for three reasons.
First: the framework producers were right to build a framework that can return N/A. This is not a trivial design decision. The overwhelming majority of analytical templates are constructed so that every cell has a default โ "likely," "moderate," "sufficient." Defaults exist because a report with empty cells is harder to sell. A framework that permits emptiness is a framework that has been designed by someone who has been burned by a filled-in report. Whoever wrote those nine dimensions has, at some point, watched a confident conclusion collapse because one row was invented. That is institutional memory encoded in a schema, and it deserves more respect than it usually gets.
Second: the projects that the bulls are excited about in this consolidation phase are, on average, more measurable than the projects of 2021. This is not a small shift. In 2021, a project's central claim was about a future state โ a roadmap, a TPS number, a partnership. In 2026, the claims that attract capital are about present state: fee revenue, net deposits, sequencer margin, unlock overhang. Those claims can be falsified with on-chain data, which means the null report's dimensions can actually be completed for them. A framework that returns N/A against a 2021 project returns real numbers against a 2026 one. That is progress, and it is progress driven by the market, not by researchers.
Third, and this is the one I find genuinely uncomfortable: the AI-generated research glut has done something useful. It has devalued prose. When a 4,000-word protocol report costs four minutes to produce, the market stops paying for prose and starts paying for verification. The null report is the natural endpoint of that repricing. It is a document that contains no prose at all โ only a list of evidentiary gaps. Ten years ago, that document would have been unsellable. Today it is more informative than the reports it replaces, because the reports it replaces are now free.
The blind spot in my own position is this. Forensic skepticism is a filter, not a generator. It can tell you with high confidence that a project's stated yield has no external revenue source. It cannot tell you which of the twelve projects with genuine external revenue will compound. The null report protects capital from the bad half of the distribution. It says nothing about the good half. Anyone who reads my work and concludes that the correct posture is to hold nothing has misread the method entirely โ they have confused the audit with the investment.
What the bulls understand that the null report cannot express: the largest returns in this asset class have historically come from the dimensions that were hardest to evidence. Bitcoin in 2013 had no revenue model that would survive a nine-dimension framework. It had a thesis. The framework would have returned N/A across eight cells and one weak affirmative in the technical dimension. The thesis was right anyway. That is not an argument for abandoning evidence. It is an argument for being explicit about what the evidence cannot cover โ which is precisely what a null report does, and precisely what a confident report conceals.
Takeaway
The null report is not a failure of analysis. It is analysis operating at its design limit, and publishing the limit rather than papering over it.
What I want to see change is narrow and specific. Every research document published in this market should carry a completed-cell count and a list of the dimensions that returned N/A. Two numbers, one list, at the top of the page. A report that says "7 of 9 dimensions completed; regulatory and tokenomic returned N/A" tells a reader more in one line than four thousand words of protocol summary.
The framework already exists. The discipline already exists. What is missing is the willingness to publish the empty cells alongside the full ones โ because an empty cell is an admission, and the industry has spent five years optimising for the appearance of completeness.
One question, then, for the next project that hands you a deck with a fully populated allocation table and a 19.5% yield.
Ask which four addresses the supply table was reconstructed from. If the answer takes more than a minute, you already have your N/A.
