A template landed in my inbox last week. Nine dimensions. Forty-one sub-headers. A confidence column, an evidence column, and an empty array where the information points were supposed to sit — zero entries, cleanly formatted, waiting for a payload that never arrived.
The pipeline blocked. Someone upstream had the discipline to refuse, and the refusal note was better written than most research I read all quarter. Here is what bothers me: it almost didn't matter. Had the gate stayed open, that template would have filled itself. Technical surface: "modular architecture, strong." Token surface: "distribution moderately concentrated, monitor unlocks." Regulatory surface: "medium risk pending clarity." Nine dimensions of confident, sourced-looking prose generated from nothing, priced at four figures, forwarded to an investment committee that would never check a single row against a block explorer.
That is not a research failure. That is the product.
The nine-dimension format spread because it looks like rigor. It has the shape of diligence — technical, token, market, regulatory, competitive, team, roadmap, risk, confidence — and the shape is what gets purchased. Nobody buys a conclusion. They buy a document that makes the conclusion defensible in a meeting. Once you understand that, the empty array stops being an anomaly and becomes the honest version of the genre: a report with no anchors, finally admitting it has no anchors.
I've watched this factory scale for two years. It used to take an analyst three days to produce a token memo. Now it takes a prompt and eleven minutes, and the output is longer. The cost of words collapsed. The cost of verified data did not. That gap — cheap prose, expensive evidence — is the entire structure of the current research market, and it is why the average "deep dive" on a mid-cap protocol contains less verifiable information than a single Etherscan tab.
So let me show you what the filled-in version is supposed to look like. Not the format. The anchors.
In late 2023 I audited EigenLayer's contracts myself before writing a word about restaking. Not because I expected to find a critical bug — I expected to understand the withdrawal queue. I found a re-entry vector in the queue logic, nothing fatal but structurally relevant, and I put $15,000 of staked ETH into the first AVS pool to test the economics with skin in the game rather than in a spreadsheet. The yield was unremarkable. The exposure was the point. When I published the breakdown, three quant firms forked it — not because the analysis was clever, but because it cited commit hashes.
That is the difference. An anchor is a thing another person can independently pull on. Auditor name plus scope plus commit hash. Unlock table plus cliff date plus percentage of float. Finality time measured on mainnet, not quoted from a docs page. If a report cannot be falsified by a stranger with a browser, it is not research. It is marketing with a bibliography.
Four fields decide whether you survive a bear market. I check them before I size anything, and I check them in this order.
Field one: audit scope, not audit name. A protocol that says "audited by [firm]" has told you nothing. Which contracts? Which commit? Were the periphery and the hooks in scope, or only the core? Uniswap V4 is the cleanest example of why this matters — hooks are programmable Lego, and the entire risk surface lives in code that the core audit explicitly excludes. A hook-level exploit does not need to touch the pool contract. When I see a V4 fork advertising a core audit and shipping third-party hooks, I mark it unhedged and I size accordingly. If the scope document is missing, the correct entry in your report is not "low risk." It is unaudited code, treat as hostile.
Field two: the cliff inside ninety days as a percentage of float. Not total supply. Not fully diluted valuation. The percentage of tradable float that unlocks within one quarter. This is the single number that has cost retail more money than every exploit combined, and it is the number most often replaced with a sentence about "long-term alignment." If the unlock schedule is unavailable, the honest entry is pending observation, no position. Not "moderately concentrated." The distinction matters because a schedule you cannot find is usually a schedule someone chose not to publish.
Field three: who actually receives the fee. In a bear market, revenue is the only thing that does not need a narrative. So trace it. Does the fee accrue to the token, to a multisig, to a foundation entity, or to a staking contract that redistributes it? I have read dozens of governance proposals where the answer was "the foundation," dressed in eleven pages of decentralization language. This is where my view on governance tokens hardened: most of them are non-dividend equity with a voting UI. The holder's only return path is a later buyer paying more. That is not a moral judgment, it is a cash-flow statement — and if the fee never reaches the token, the report should say so in the token dimension, in plain language, not in a footnote about "value accrual mechanisms under discussion."
Field four: who can change parameters overnight. Pull the multisig signer list. Count the keys. Check whether the timelock is real or a two-hour courtesy window. Governance capture is not a theoretical risk — it is a Tuesday. A protocol where four addresses can reprice collateral or pause withdrawals is a protocol whose risk model has four humans in it, and your report should name them.
Those four fields are the technical and token dimensions. The market dimension is where I will accept external data, because I can pull it myself: TVL drawdown over thirty days, funding rate sign and magnitude, open interest relative to float, and the unlock calendar forward. But even here the discipline holds — label what you fetched, label what you inferred, never blend the two. When I model Layer 2 fee revenue, I do not use today's blob price. Dencun made blobs cheap and every rollup built a business model on that subsidy; blob demand is climbing against a fixed supply schedule, and when it saturates, rollup gas costs step up hard. Any L2 valuation that assumes current data costs is a valuation with an expiration date printed on it. That belongs in the report as a stated assumption, not buried in a model tab.
Here is the contrarian part, and it is uncomfortable.
The empty array was the most valuable artifact in that whole pipeline. A blank field is a signal — it tells you exactly where the seller of the report had no leverage, no access, no verified data. The blanks are the alpha. Everyone reads the conclusions; almost nobody reads the gaps. I have made more money from the missing rows than from the filled ones, because a report with three honest blanks tells me precisely which questions the market has not priced. A report with zero blanks tells me the author filled them with narrative, and narrative is the cheapest input in this industry.
Which brings me to the demand side. These reports are not produced to inform decisions. They are produced to defend decisions already made — by a fund that needs an IC memo, by a KOL who needs a reason for a bag, by a team that needs third-party credibility. The template exists because the buyer wants the shape of diligence without the cost of it. That is a market, and it is clearing, and it will keep clearing until the next blowup forces everyone to read the footnotes.
In the sprint, hesitation is the only real cost. But there is a second cost nobody prices: the hesitation to write "insufficient data." Every fabricated confidence score is a deferred loss, transferred from the analyst who should have admitted ignorance to the reader who acted on it.
So here is what I want from the next report you read, and from the next one you write. Three thresholds, checked before any position: unlock cliff above five percent of float inside ninety days — pass. Fee capture not reaching the token — pass. Audit scope excluding the code that actually custodies funds — pass. Three passes, or no size. Not a smaller size. No size. The blank field is not a gap to be filled with judgment; it is an instruction to stand down.
The next cycle's edge will not be better models. Everyone will have the same models. It will be better blanks — analysts disciplined enough to leave a row empty, and readers disciplined enough to notice which rows those are. The template that arrived in my inbox last week refused to fabricate. That should be the floor of this industry, and right now it is the ceiling.
