I opened the file on a Tuesday morning in February, coffee going cold beside the keyboard, and I remember the exact second my stomach dropped.
Forty-one pages. Executive summary. Technical architecture diagram with clean vector lines. Token distribution pie chart. Risk matrix in three colors — green, amber, red, the palette of a traffic light suspended over an empty road. Ecosystem dependency map. Regulatory posture table. Nine analytical dimensions, each with its own typography, its own sub-headers, its own conclusion box.
Every conclusion box said the same sentence.
N/A — insufficient information.
The pie chart had no slices. The risk matrix had no cells filled. The dependency map had three boxes and all three of them said N/A. Somebody had spent real hours formatting emptiness. It was the most beautifully typeset void I have ever paid for.
Four thousand dollars. Three weeks. And it wasn't fraud. That's the part that has kept me awake since.
The analyst wasn't lazy. The analyst was structurally prevented from doing the work, and the template would not let them say that in plain language, so it made them say it forty-one different ways across nine branded dimensions instead.
Let me tell you what happened in that engagement. It was a modular data-availability layer with a token that had 68% of supply still locked, a mainnet eleven weeks old, and a team that had published two blog posts in six months. The analyst asked for the cost basis on the sequencer infrastructure. No answer. Asked for the unlock schedule as a table. Got a link to a tokenomics post with an illustration of a rocket. Asked for the audit report. Got a single page with a logo and the line "no critical findings."
Every question landed in the same place. Not refused. Just unanswered. And the template had no field for "the team declined to answer," so the answer became N/A, and the N/A became a page, and the page became a report, and the report went to an investment committee that read the executive summary, saw nine dimensions addressed, and moved on.
Nobody lied. The system lied for them.
That is the mechanic I want to spend the next several thousand words on. Not "crypto research is bad." Everyone knows that. The specific, mechanical thing — how structured templates convert the absence of information into the appearance of presence — is what is actually costing money right now, in a market where capital is scarce and the difference between a surviving protocol and a dead one is three months of runway.
So let me audit the audit. Dimension by dimension. Seam by seam.
A short history, because the templates did not appear from nowhere.
Between 2017 and 2019, crypto research was vibes with charts. A blog post, a Telegram group, a founder on a podcast, a spreadsheet with prices in it. The ICO era ran on whitepapers written in LaTeX and read by nobody. I was there for that. I co-hosted a podcast called Chain of Thought, and we interviewed twelve founders. What struck me, over and over, was that the projects with the thinnest technology had the thickest documents. The whitepaper was the product.
Then 2020 happened. DeFi Summer. Suddenly there were numbers — TVL, APR, liquidity depth, utilization. And with numbers came scorecards. By 2021 every fund had a framework. Nine dimensions, ten dimensions, 1-to-5 ratings, weighted averages, a methodology slide. The methodology deck became a fundraising asset in itself. I watched a fund raise on the strength of its "proprietary research stack" before it had deployed a single dollar.
Then the ETF era arrived. In early 2024 spot Bitcoin ETFs were approved, and the audience changed. The people consuming research were no longer crypto natives who could smell a dead protocol through a screen. They were allocators with fiduciary duties, compliance officers, risk committees. They wanted documents. They needed documents, because their internal processes demanded artifacts, and you cannot staple a Telegram message to an investment committee memo.
We didn't build a research culture. We built a research aesthetic. And we scaled it at exactly the moment the stakes went up.
Here is the arithmetic nobody runs. A template has a fixed number of fields. The supply of honest, verifiable information about a crypto protocol in a bear market is roughly constant and very low. Force the second through the first and you do not get more information. You get one of three outputs: padding, fabrication, or N/A.
Padding is the most common. The analyst writes four hundred words about an architecture they do not understand, sourced from the project's own documentation, cited as "team materials." That is not analysis. That is transcription with a byline.
Fabrication is rarer, but it is real. Estimates presented as measurements. "Annualized revenue" derived from a two-week fee sample collected during an airdrop campaign. The sample is real. The annualization is a lie.
And N/A is the honest one. It is also the one that costs the analyst their next engagement.
I have filled in those templates. Based on my audit experience across eleven protocol codebases in four years — three rollups, two restaking layers, six applications — I have written the N/A more times than I have written anything else. I once had an engagement lead ask me to "find a way to say something" about a governance model where a foundation wallet controlled 61% of voting power and turnout sat under 2%. There was something to say. It just wasn't what the client had purchased.
So. Nine dimensions. Let's find the seams.
Dimension one: technical.
Start here, because it is where the template is most confident and least useful.
The technical section of a standard framework asks for innovation, maturity, security assumptions, performance. Four rows. Clean. Comparable. Utterly fictional in practice.
Innovation is not measurable. Maturity can be measured, but nobody wants the number, because the number for most production rollups in 2026 is somewhere between eighteen and thirty-six months of unbroken operation. And the interesting question is not how long the chain has run. It is how many times it has been forced to run without a human intervention. Almost none. Every L2 in my sample had, at some point in the previous year, required a manual sequencer restart or a forced upgrade through an admin key. That is not a criticism of the teams. That is the actual state of the art, and it should be printed in bold on the cover of every report.
But here is where the template does real damage. It asks for security assumptions as a row, and the analyst writes something like "trusted setup plus fraud proof window." Which is true, and which tells you nothing about whether the thing works.
Let me give you the specific that actually matters, and that the template has no field for.
A zero-knowledge rollup's cost structure does not move with its revenue structure, and that divergence is the whole story of 2025 and 2026.
Post-EIP-4844, blobspace made L2 data availability cheap. Blob fees collapsed to a fraction of calldata prices, and in many periods the blob fee market cleared near the minimum. L2 transaction fees fell accordingly — by 80% to 95% at the median across the major general-purpose rollups. Users celebrated. Analysts wrote "improved unit economics."
But proving costs did not fall. A ZK rollup must generate a validity proof for every batch it settles, and that proof is produced by a prover running on specialized hardware — GPUs, increasingly FPGAs, occasionally ASICs from two or three vendors. That hardware has a depreciation schedule. It has a power bill. It has a pipeline that must keep producing proofs whether or not the network is busy, or the latency between blocks stretches past what users will tolerate.
So the fixed cost sits there. And the variable revenue — the fee the rollup collects, minus the blob cost it pays — got compressed from the top.
I have looked at three ZK proving operations in the last eighteen months. At the low end, a mid-sized ZK rollup is spending somewhere in the low-to-mid six figures per month on proving infrastructure once you amortize hardware and include the engineering cost of keeping the prover pipeline alive. Pre-4844, that was recoverable. Post-4844, with fees down 90%, it is not recoverable from fees. It is recoverable from a token treasury, or from a venture round, or from a sequencer margin the chain is quietly subsidizing and not disclosing.
The number the template wants is "performance." The number that matters is cost per proof. No team publishes it.
And the template has no row for "the team declined to publish cost per proof." So the analyst writes N/A, and the N/A reads like an administrative gap rather than what it is, which is the single most important financial fact about the business.
The pivot wasn't from optimistic rollups to ZK rollups because the technology got dramatically better. The pivot happened because ZK became fundable, and the padding followed the money.
Dimension two: tokenomics.
This is the dimension where the template is at its most seductive, because tokenomics has numbers. Tables of allocations. Cliff dates. Vesting curves. It all looks rigorous.
And it is the dimension where the N/A is most often a lie by omission.
Here is the standard table, filled in as best as anyone can: team, 20%, twelve-month cliff, forty-eight-month linear vest. Early investors, 25%, twelve-month cliff, twenty-four-month linear. Community and liquidity, 30%, various. Treasury, 15%, controlled by foundation, schedule unspecified. Public distribution, 10%.
Clean. Standard. And it tells you almost nothing, because the number that determines whether a token survives a bear market is not the allocation percentage. It is the circulating float against the unlock wall, and the identity and cost basis of the counterparties on the other side of the cliff.
Let me be concrete. If a token has 12% circulating supply and a 25% investor allocation with a twelve-month cliff landing in a market where daily spot volume has fallen 70% from its peak, that cliff is not a "supply event." It is a liquidation event with a countdown timer. The investors' cost basis is not the ICO price. It is a fraction of it. They are not deciding whether to sell at a loss. They are deciding whether to sell at forty times or twenty-five times.
The template asks for allocation. It should ask for cost basis dispersion. It never does.
And here is the second thing it misses, and I want to be careful with this, because it is the point where most frameworks retreat into a comfortable narrative.
"Liquidity fragmentation" is not a technical problem. It is a funding thesis wearing a technical problem's clothes.
Every cycle, some group of well-capitalized people decides that the real issue in DeFi is that liquidity is scattered across too many venues, and the solution is a new layer — an aggregator, a solver network, an intent layer, a chain-abstraction protocol. Each of these is a real piece of engineering. Each of them also requires a token.
Look at where the liquidity actually is. In stablecoin pairs on two or three venues. In liquid staking tokens on two or three more. In a handful of perpetual markets. That is not fragmentation. That is concentration, and concentration is not a bug you fix with a routing layer. It is what happens when liquidity providers put capital where the volume is, and volume goes where the liquidity is, and the equilibrium settles at three or four venues and stays there.
You do not need a new protocol to fix that. You need to be honest that the equilibrium is efficient and that the "problem" is a market that has already decided.
I watched a company raise a seed round in 2024 on a deck whose central slide was a map of DeFi with a hundred dots on it and the word "fragmented" in the corner. The map was accurate. The inference was not. A hundred venues holding a thousand dollars each is not fragmentation. It is dust. And you cannot route dust into liquidity.
The template has a row for "competitive advantage," and the honest entry for most of these protocols is "none, the market does not want this." Nobody writes that. They write "first mover in intent-based settlement." Which is a sentence, not an advantage.
Dimension three: market.
The market section is where the bear market does the analyst's job for them, and where most frameworks still get the sign wrong.
Funding rates are the most honest number in crypto. Not price. Not TVL. Funding. Because funding is a payment — an actual transfer of value between actual counterparties, settled every hour on perpetual futures venues, and it cannot be faked with a screenshot. When funding is persistently negative on a major asset, someone is paying real money every hour to be short. That is conviction with a cost attached.
In this cycle, funding on the majors has spent long stretches negative, which means the perp market is paying longs to hold. In 2021 the same number was the mirror image, and everyone read it as confirmation. Nobody reads it as confirmation now. That, at least, is progress.
The other honest number is stablecoin net issuance. Stablecoin supply is the closest thing crypto has to a money supply aggregate, because stablecoins are the settlement asset. You cannot buy a token without stablecoin liquidity somewhere in the chain, and you cannot exit without creating demand for it. In a healthy market, stablecoin supply expands. In a contracting market, it flattens or shrinks, and the shrinkage leads the price, not the other way around.
Right now the supply is roughly flat with episodic outflows during liquidation cascades. That is what a bear market looks like at the plumbing level. Not a crash. A slow draining.
And the template asks for none of this. It asks for "price impact assessment" and "overall sentiment," both of which are unfalsifiable, and both of which the analyst will fill with a paragraph of hedging because there is no way to be right or wrong.
Dimension four: ecosystem niche.
Upstream dependencies. Downstream integrators. Developer signals. User signals. It reads like a supply chain audit, and it is the section most likely to be filled with numbers that are technically true and practically meaningless.
Contributor counts are gameable. GitHub commits are gameable — a bot that reformats a README four times a week produces a commit graph indistinguishable at a glance from active development. Contract deployment counts are the most gameable of all, because on most chains you can subsidize them. I have seen grants programs where the metric of success was "contracts deployed," and the result was a few thousand identical ERC-20 clones deployed by the same dozen wallets, each one collecting a small incentive.
Retention is the only honest metric in this section, and nobody publishes it.
Here is what retention actually looks like in a bear market. Take a protocol that ran an incentive campaign eighteen months ago. At the peak it had a hundred thousand weekly active addresses. Six months later, with the incentives wound down, it has eleven thousand. Twelve months later, it has four thousand. The four thousand are real — they are people who found a product rather than a payout. That number, four thousand, is the entire value of the protocol, and it is the number the team will never put in a deck, because four thousand is smaller than the seed round's implied TAM.
So the report says "strong community engagement" and moves on.
Dimension five: regulation.
I want to defend this N/A, because it is the one place the template's honesty is actually correct.
The Howey test has four prongs: investment of money, common enterprise, expectation of profit, and reliance on the efforts of others. The template asks the analyst to assess each. The analyst writes N/A on all four.
And that is the most accurate legal opinion ever printed, because the actual state of the law in 2026 is that a token's status depends on the facts and circumstances, the distribution method, the marketing language, the degree of decentralization at the moment of sale, the jurisdiction of the purchaser, and the current posture of whichever regulator happens to be paying attention that quarter.
I have sat in rooms where three lawyers gave three different answers about the same token and all three were defensible. That is not a failure of the lawyers. That is the actual condition.
But the template converts this genuine ambiguity into an apparent gap in the analyst's work, which is backwards. The correct output is not N/A. The correct output is: this asset is legally undefined, and undefined is a risk category, not a data gap.
Here is where I think the industry got something right without meaning to. On-chain settlement is now the primary record. Not the marketing page, not the term sheet, not the safe agreement. The transaction. Trust is no longer a promise; it's a protocol — and the protocol doesn't care what your counsel memo says. It executes what it was told to execute, and the enforcement question becomes a question about the chain rather than about the memo.
That is a real shift. It is also the reason compliance teams are so uncomfortable, because the artifact they used to rely on — the document — is no longer the authoritative record.
Dimension six: team and governance.
Anonymity. Turnout. Concentration. Proposal quality.
Governance is where crypto's aspirations and crypto's practices are furthest apart, and the template makes it worse by treating it as a checkbox.
Here is a real governance snapshot from a protocol I will not name. Total token supply: one billion. Eligible to vote: 340 million, roughly, based on the snapshot at the proposal block. Quorum requirement: 4% of eligible supply. Votes cast: 2.1% of eligible supply. The proposal passed because a foundation-controlled multisig voted, and the foundation held enough tokens to clear quorum on its own.
That is not a governance attack. Nothing was hacked. Everything followed the rules. The rules simply permitted a single entity to be the quorum.
And here is the part that leaves a mark on you. The proposal had a discussion thread with ninety-four comments. Genuine ones. People had read the code, they had opinions, three of them had found an actual bug in the upgrade path. The vote was a formality. The discussion was where the governance actually happened, and the discussion had no binding force whatsoever.
A framework that asks "what is the governance model?" will get the answer "token-weighted, on-chain, with a four percent quorum." Every word of that is true. None of it tells you that a single wallet can decide.
Dimension seven: risk.
The risk matrix is where the template's failure becomes existential, because the risk matrix is the artifact people actually read.
Six categories. Technical, market, operational, regulatory, competitive, narrative. A rating, a probability, an impact, a mitigation. It looks like a quantitative instrument. It is a mood board.
And there is a deeper problem. The largest risk in any crypto position taken in 2026 is the risk that the analyst filling the matrix is filling a matrix.
I mean this literally. If the process that produced your risk assessment was a template that had to be completed regardless of available information, then the risk assessment contains no information about risk. It contains information about the template. And a committee that reads it and approves an allocation has not evaluated risk. It has evaluated formatting.
There is a name for this in safety engineering. It is called a normalization of deviance — when a system's output looks correct for long enough that nobody checks whether the underlying process still works. Crypto's diligence process has been producing good-looking artifacts for three years. Almost nobody has checked whether the artifacts correspond to anything.
Dimension eight: narrative.
Narrative is the dimension analysts most enjoy and least quantify. And in a bear market, narrative is not decoration. Narrative is revenue.
Let me give you the case I keep coming back to, because it is the cleanest one in the entire space.
For most of Bitcoin's post-2020 life, blockspace was not scarce. Blocks were not full. Fee revenue was a rounding error against the block subsidy, and the subsidy halves. April 2024 took it from 6.25 to 3.125 bitcoin per block. The next halving takes it to 1.5625. The security budget — the total amount paid to miners to keep the chain running — is converging on the fee market, and the fee market, for a decade, was empty.
Then inscriptions arrived, and with them a wave of demand for block space that nobody had modeled. Ordinal theory gave a reason to put arbitrary data in witness space. Inscriptions, then BRC-20, then Runes. Fee revenue spiked to levels not seen since 2017. During peak periods, fees constituted a substantial share of total miner revenue, briefly exceeding the subsidy in some blocks.
I have argued this before and I will keep arguing it, because the implications are structural. Without the inscription wave, Bitcoin's security model would already be in visible trouble. Not theoretical trouble. Numbers trouble. When fees are 1% of miner revenue and the subsidy is cutting in half every four years, the chain is on a schedule to underpay its own security.
The institutional adoption of 2024 and 2025 — the ETFs, the corporate treasuries — did not solve that. It raised the price of the asset, which raised the dollar value of the subsidy, which bought time. It did not create blockspace demand. Blockspace demand comes from people who want to put data on Bitcoin, and for a moment, they showed up.
And here is what the template does with this. It has a row called "narrative sustainability" and the analyst writes "strong, driven by ETF flows." Which is true and useless. The question that matters is: what is fee revenue per block in a week when nobody inscribes anything? And the answer, for most of 2025, was: low enough to be scary.
Dimension nine: supply-chain transmission.
Last dimension. Where a failure in one layer becomes a failure in another.
In 2022 this was easy to watch, because the transmission was fast. Terra's collapse drained the funds that had been farming Anchor. Those funds were leveraged. The leverage was held by Three Arrows Capital. Three Arrows' failure forced liquidations that hit Celsius, which froze withdrawals, which triggered a wave of miner capitulation as the collateral behind equipment loans got called.
Four links. Six weeks.
The 2026 transmission is slower and more interesting, because it runs through margin rather than through collateral. Start with the L2 fee compression I described earlier. That compresses fee revenue for the rollups. That compresses the sequencer operators' margins. A compressed sequencer margin makes the operator more dependent on token emissions to pay for infrastructure, which makes the token the funding source for the prover pipeline, which means the redemption schedule of that token is now a systemic input into the chain's availability guarantees.
Now add the DA layer, which is priced against blob demand, which falls when L2s post fewer batches. Add the proving market, where a handful of prover operators serve most of the ZK rollups, which makes them a single point of failure with a corporate balance sheet attached. Add the ETH burn, which collapses when L2 settlement posting drops, which removes a supply sink that a lot of valuation models assumed was permanent.
None of these links is dramatic on its own. Together they describe an ecosystem where the revenue that funds the security is shrinking faster than the cost structure that delivers it. That is what bleeding looks like. Not a gap down. A slow margin squeeze through four layers of infrastructure, most of which don't publish their financials.
And the template has a row for "downstream impact" that the analyst fills with three sentences because the engagement ended last Thursday.
The fabrication pressure.
I have described nine dimensions. I have not yet described the thing that makes the nine dimensions fail, which is the incentive structure around the analyst.
Here is how it works in practice. An analyst at a research firm or a fund is evaluated on output. Number of reports. Number of protocols covered. Depth of coverage. Turnaround time. None of those metrics distinguish between a report that contains information and a report that contains formatting. Both count as one.
So the analyst produces the report. And when the information isn't there, they have three options. They can write N/A and produce a report that looks thin and gets them a conversation with their manager about why they're not adding value. They can pad, and produce a report that looks thorough. Or they can estimate, and produce a report that looks rigorous.
Two of those three options get you promoted.
The engagement lead has the same problem one level up. The client is paying for a report. A report that says "we cannot assess this" invites the client to wonder what they paid for. A report that says "nine dimensions, all covered" does not.
And the client — the allocator, the committee, the LP — has the same problem again. They need an artifact for their file. The N/A report is an artifact that says "we looked and found nothing," which invites a follow-up question they cannot answer. The padded report says "we did the work."
The whole chain, from analyst to LPs, is optimized to produce the appearance of diligence rather than diligence. And because everyone in the chain understands this implicitly, nobody pushes back, and the equilibrium holds.
I learned to stop preaching and start listening on this one. For years I wrote about how the industry needed better research standards. Standards were not the bottleneck. Incentives were. You can write the best methodology in the world and it will be gamed by anyone whose bonus depends on filling it in.
AI agents and the N/A problem.
There is a new layer now, and it makes everything I have described faster and worse.
In 2026, autonomous agents are querying on-chain data, composing reports, and in some cases executing allocations. Agent wallets hold balances. Agent frameworks read subgraphs and produce summaries. You can point one at a protocol and get a nine-dimension assessment in forty seconds.
I built part of this. My organization runs a research stack that uses agents to pull on-chain state, and I have watched an agent produce a complete, confident, beautifully structured analysis of a protocol whose mainnet had been paused for three days. The agent did not know the chain was paused. Nothing in its data sources said so. So it filled the dimensions, and it filled them well, and a human who skimmed it would have nodded.
Code is law, but empathy is the interface. The agent could read every state variable on that chain and still not understand that a paused chain means people cannot move their money, and that a person who cannot move their money is not a data point.
This is the part of the Human-Centric Blockchain work that keeps me going. The question is not whether agents can produce analysis. They can. The question is whether the analysis verifies human intent, or merely optimizes for the appearance of verification. An agent that fills nine dimensions is doing exactly what the analyst was doing, only faster, which means the failure mode now scales with compute.
I keep coming back to a simple test. Before you accept any assessment, ask what the assessor would have had to say to earn their money if the answer had been "we don't know." If there is no such answer available to them, the assessment is not information. It is a receipt.
Here is the counter-intuitive part, and it's the thing I want you to take with you, because everything above is a diagnosis and this is the actual argument.
A report where every field is green is more dangerous than a report where every field is N/A.
I mean that as a hard rule, not a rhetorical flourish. In an industry where verifiability is structurally limited — where teams can decline to answer, where financials are unaudited, where governance can be quorum'd by a single wallet — a completed assessment is a claim of knowledge. And a claim of knowledge under those conditions is either naive or it is a sales document. There is no third option.
The N/A report is honest. It is also useless. And that combination — honest and useless — is not what people want to buy, which is exactly why it gets replaced.
So the question is not "how do we get better reports." The question is: what would diligence look like if it stopped pretending to be a document?
Here is the shape of it, as best as I can see.
Fewer dimensions, more questions. Four questions instead of nine dimensions. Who pays for the prover? Who is on the other side of the unlock? What is fee revenue in a week when nobody is incentivized to transact? And who can pause the chain? If you can answer those four, you know more than any nine-dimension scorecard will ever tell you.
Verifiable over attested. Anything that can be checked on-chain should be checked on-chain, at the moment of the decision, not transcribed into a table three weeks earlier. The moment you accept a PDF as evidence, you have accepted a trust relationship. Trustless systems require trusting relationships — and we have spent three years laundering trust through documents instead of building processes that don't need it.
And price the N/A honestly. If a team declines to disclose cost per proof, that is not an administrative gap. That is a disclosure event, and it should carry the same weight as a qualified audit opinion. Right now it carries zero weight, because the template has nowhere to put it.
I have watched this industry build extraordinary infrastructure for verifying transactions. Immutable ledgers. Cryptographic proofs. Solvers that settle atomically. And then I have watched the same industry verify the parties it funds with a PowerPoint and a vibe.
That gap is the actual bear market problem. Prices recover. Margins recover. The habit of accepting a well-formatted void as evidence does not recover on its own. It just gets inherited by the next cycle, with better fonts.
The next time someone hands you a report with nine dimensions and a green risk matrix, ask them one thing. Ask them what the report says in the places where they didn't have an answer.
If they can't tell you, you haven't received an analysis. You've received a template, and the template has been lying to you with your permission.
