Last Tuesday I ran a nine-dimension diligence framework against a restaking-adjacent rollup that closed a nine-figure round in the spring. I expected friction. I got blanks.
Seven of the nine fields returned nothing verifiable. Token supply schedule: not disclosed. Revenue source: "protocol fees." Jurisdictional footprint: unstated. Team: partially anonymous. Unlock curve: "TGE plus linear." Competitor TVL: no comparison published. Governance timelock: referred to as "progressive decentralization." Two fields survived โ a testnet transaction count and a Twitter follower gradient โ and both of those I could have generated myself in an afternoon.
The report came back clean. That is precisely the problem with clean reports.
I have spent sixteen years reading market structure, and the last stretch of it on a 7x24 surveillance desk where the job is not to explain a move after the candle prints. Surveillance isn't prediction โ it's anticipating the break before it happens. And the break almost never announces itself in a headline. It announces itself in an empty field, in a footnote that says "to be determined," in a chart whose y-axis has no units.
Why the checklist is the most dangerous instrument in a bull market
Markets like this one do not reward skepticism. They penalize it, briefly and publicly. Every funding round that closes at a valuation you find indefensible is a data point that says your model is wrong, and enough of those data points will wear down even a disciplined analyst into accepting the deck at face value.
So the industry replaced judgment with templates. Nine boxes. Tick them. Ship the memo.
The template itself is not the failure. I built my own version of this in 2017, and it was the most useful thing I ever made โ not because it told me what was safe, but because it refused to tell me anything at all when the inputs were missing. That refusal is the entire product. A framework that produces a rating from zero inputs is not a framework. It is a mood.
The template I ran last week arrived with the standard scaffolding: technical positioning, token economics, market structure, ecosystem placement, regulatory exposure, team and governance, risk matrix, narrative and expectation gap, and supply-chain transmission. Nine dimensions, and each one requires a minimum input before it is permitted to speak. Technical positioning needs a layer, a comparison metric, and a development stage. Token economics needs supply, float, allocation, and the source of the incentive. Risk needs at least one implementation detail and one asset-flow figure. Without those, the honest output is not a low score. It is a blank.
There is a difference between one star and N/A. One star means the content exists and it is bad. N/A means the content does not exist, and the correct response is not to discount the asset. It is to treat the absence as the primary finding.
Here is the operational principle I have run on since the HotCo sprint: the blank field is not a gap in your research. It is a gap in their disclosure. Those are different problems, owned by different parties, and the second one is almost always deliberate.
The four dimensions that actually break projects
Nine sounds thorough. In practice, four of them kill positions. The other five shape sizing and timing. I will go through the four with the specificity the template demands, because abstraction is where bad diligence hides.
One: the supply arithmetic, and the cliff nobody modeled
The single most common failure in the current cycle is a project that is genuinely good and a token that is structurally incapable of holding value, because the float is a rounding error against the fully diluted number and the unlock wall lands inside the next two quarters.
| Metric | What the deck says | What the chain says | |---|---|---| | Circulating supply at TGE | "Community-first distribution" | 11.4% of total, of which roughly half is liquidity provisioning | | Fully diluted valuation | Not printed | 92x current annualized protocol revenue | | Team allocation | "Long-term aligned" | 19%, 12-month cliff, then monthly | | Investor allocation | "Strategic partners" | 27%, cliff expiring 30 days after TGE | | Emission source | "Sustainable" | 100% inflationary; no fee capture to the token |
The math is not complicated, which is why its absence is informative. A 27% allocation whose cliff expires one month post-listing is not a strategic partnership. It is a scheduled sell program with a branding budget. The discount at which that supply clears is not a discount to fair value. It is a discount to the price the market was told to expect.
My rule on floats: I do not model the price. I model the ratio of new sellable supply to organic daily volume for each of the next eight quarters. If any quarter exceeds 1.5x the trailing thirty-day average volume, the token has a mechanical headwind that no narrative can outrun. That is not a forecast. That is arithmetic with a calendar attached.
Two: where the yield actually comes from
This is where the current cycle is most dishonest, and it is where my 2020 arbitrage work still pays rent.
Back then I mapped Uniswap pool incentives against Compound's lending rates and found a spread that existed only because two systems priced the same risk differently. That spread was real. It was also temporary, and the moment two hundred traders read the same note, it was gone. The lesson I carried out of that summer was not "yield farming works." It was that yield has an address, and you can look up who is paying.
The lending markets that anchor most of this cycle's leveraged yield are not price-discovery mechanisms. They are governance-parameterized curves. Aave's rate model bends around a utilization kink that sits at a threshold someone voted on. Compound's jump-rate model does the same thing with a multiplier set by a proposal, not by a clearing auction. These constants are administrative. When utilization crosses the kink, the borrow rate does not rise because the market demanded it โ it rises because the curve was drawn that way in a forum post.
That matters because it means the yield curve on-chain is not a signal about scarcity of capital. It is a signal about which parameters are currently in force. And when a new protocol advertises a double-digit stablecoin yield, the first question is not "how." It is "from whose balance sheet."
Yield is the bait; liquidity is the trap. If the yield is funded by emissions, the depositor is being paid in the same instrument they are being asked to hold. If it is funded by points, the depositor is being paid in a promise that converts at a rate set by the issuer after the fact. Both structures work beautifully while inflows exceed outflows. Neither survives the first week where they do not.
The audit-grade test: take the advertised annualized rate and subtract the portion sourced from token issuance, points, and third-party subsidy. What remains is the real yield. In most of the pools I have examined this quarter, the remainder is below 30% of the headline, and in several cases it is negative once you mark the reward token to market at the unlock-price implied by the emission schedule rather than the current spot.
Three: the cost curve underneath the rollup
Every L2 pitch I have read this year contains the same line: fees are near zero and will stay there. That line was true for roughly the eighteen months following Dencun, and it is expiring.
Blob space is a metered resource with its own fee market. The target sits at a fixed number of blobs per block, the maximum sits above it, and when rollups collectively push past target, the blob base fee adjusts upward block by block until demand clears. That mechanism is elegant and it is unforgiving, because every rollup on the market is competing for the same fixed quantity of data availability, and every one of them has a business model predicated on that quantity growing faster than demand for it.
The pressure is already visible in the governance record. The blob target has been raised once since introduction, and the argument for raising it again is being made by the same teams whose margins depend on it. A resource whose capacity is expanded whenever its largest consumers complain is not a market. It is a subsidy with a committee.
My working model is that blob demand reaches the new ceiling within two years, and when it does, the marginal cost of data availability for every rollup reprices at once. Not gradually. In blocks. The base fee for blobs can move from a nominal floor to a figure three orders of magnitude higher across a short sequence of blocks, and any rollup whose fee logic assumes a floor is running an unhedged short on a volatile input.
Watch the sequencer margin, not the user fee. Most rollups will eat the increase for a quarter or two to protect the growth chart, then pass it through. The chart they were protecting is the one that drew the depositors.
While I am here, the same logic explains why I do not allocate risk budget to Bitcoin programmability experiments. Pushing fungible-token standards and inscription metadata through a chain designed for settlement is using a Rolls-Royce to haul cargo โ it insults the vehicle and it does not carry much. The engineering is clever. The throughput and the fee dynamics are not, and every cycle someone rediscover this at the cost of a position.
Four: governance is a timelock and a key
The governance dimension is the one most often answered with a word salad. "Progressive decentralization" is not a governance model. It is a schedule without dates.
Three facts settle this dimension in under ten minutes. Does an upgrade execute behind a timelock, and how long is it? Who holds the sequencer, and can it reorder or censor? Is the proxy admin a multisig, and what is the threshold?
| Configuration | Timelock | Sequencer | Proxy Admin | Verdict | |---|---|---|---|---| | A | 0 hours | Single operator | 2-of-3, unnamed | Uninvestable at size | | B | 48 hours | Single operator | 4-of-7, named entities | Positional, size-limited | | C | 7 days | Decentralized set | Governance-executed | Acceptable | | D | "Soon" | "Soon" | "Soon" | Blank field. Treat as A |
Configuration D is the default in this cycle, and it is the default for a reason. A timelock longer than the market's reaction function removes the ability to front-run the team. A timelock short enough to be operationally convenient removes the protection entirely.
I watched this failure mode up close in 2017, auditing early ERC-20 contracts during the sprint that made my name. We found an integer overflow in a token contract that could have drained roughly two million dollars of user balances. The code had been reviewed. It shipped anyway. The lesson I took from it โ and the reason I still read bytecode before I read the blog post โ is that "audited" is a claim about a date, and a date is not a property of the code. The vulnerability existed after the review, because the review did not cover the arithmetic path the attackers used. Similarly, "decentralized" is a claim about a roadmap, and a roadmap has never censored a transaction.
The contrarian read: the blanks are the product
Here is where I diverge from the desk consensus. The standard interpretation of a diligence report with seven blanks is that the analyst did not work hard enough. The standard remedy is to dig until the fields fill in.
I think the blanks are frequently designed, and digging is often exactly what the issuer wants you to do.
Consider what ambiguity buys. An undisclosed unlock curve preserves optionality for the team and removes it for you. A vague revenue description allows the token story to be told two ways in two jurisdictions. A partially anonymous team lets the market price the deliverables rather than the rรฉsumรฉ, right up until the moment the deliverables slip and there is no one to be accountable. None of this is accidental. It is the same logic as a deadline without a penalty: the party who set the terms keeps the option.
So the framework's real output is not a rating. It is a map of where the price is unsupported. Every blank field is a coordinate where sentiment is carrying the valuation alone, and sentiment has a shorter half-life than supply. The price is a reflection of sentiment, not value โ and when seven of nine inputs are missing, the sentiment share of the price is roughly the whole thing.
There is a cost to running this discipline in a cycle like this one, and I will not pretend otherwise. Demanding inputs before sizing means missing the first forty percent of a move, repeatedly, while people who read nothing outperform you in public. That is the negative carry of rigor. It resolves at the point where the unlock calendar arrives, or the blob base fee reprices, or the timelock turns out to be zero. A red candle doesn't lie โ but it lags. By the time it prints, the mechanics that caused it were visible in a document nobody finished reading.
What I am watching next
Four triggers, in order of signal value.
Unlock calendars landing inside a quarter whose sellable-supply ratio exceeds trailing volume by more than 1.5x. Blob base fee sequences that hold above target for consecutive epochs โ that is the tell that the subsidy era is ending, and rollup fee schedules will follow. Timelock upgrades from below 48 hours to anything longer, announced quietly alongside a partnership. And the quiet reissuance of a token's disclosure document after a funding round, where fields that were blank last quarter suddenly have numbers โ which tells you the numbers existed all along and were being released on a schedule.
Don't fight the tide. Map it. The blanks will fill in eventually. The question is whether they fill in while you are holding, or while someone is selling to you.