A deal memo landed on my desk in March. Series A. One hundred million dollars raised. Post-money valuation, one point two billion. The syndicate page listed four funds I respect and one I do not.
I ran standard intake. Nine dimensions. Technical architecture. Token economics. Market structure. Ecosystem position. Regulatory exposure. Team and governance. Risk matrix. Competitive set. Data provenance.
What came back was a form where every cell carried the same string: N/A — insufficient information.
Not one technical specification. Not one supply table. Not one contributor count. Not one commit in the last ninety days. The upstream analyst had done the correct thing and refused to fabricate. Downstream, in the deal tracker, someone had already typed 'conviction: high.'
The most expensive word in crypto research is not 'risk.' It is 'approximately.' Approximately decentralized. Approximately audited. Approximately five percent of supply circulating. Each 'approximately' is a null input that somebody converted into a number, and every converted null is a position size set on an assumption rather than an observation.
That memo is the artifact I keep returning to, because it is not an edge case. It is the default condition of this market. We are in a bull market. Capital is moving fast. Timelines are compressing. And the research function inside most funds has quietly been reorganized from a verification department into a narrative department. The mandate changed. Nobody wrote it down.
This brief is about the null input. What it is. Why your pipeline converts it into conviction. And how to wire a circuit breaker before it converts into a loss.
Context: How Research Became a Supply Chain of Inference
A modern crypto research note is a supply chain. Raw feeds enter at one end: on-chain data, repositories, governance forums, Discords, founder interviews, social threads, exchange announcements. A position exits at the other end: long, short, pass, or size-up.
Somewhere in the middle sits a machine that turns inputs into outputs. In a well-run shop, that machine is a compiler. It takes verifiable primitives and produces a decision. It fails loudly when a primitive is missing. It does not guess. It does not interpolate. It halts.
In most shops, that machine is not a compiler. It is a language model with a sales quota. When a primitive is missing, it does not halt. It fills. It fills from narrative. It fills from the last project that looked similar. It fills from the founder's tone of voice, the polish of the deck, the fact that four respected funds are already on the cap table.
Missing data does not stay missing. It gets filled — and the filler is always the most available narrative, not the most accurate one.
The incentive structure guarantees this outcome. Analysts are paid for coverage, not for refusal. A note that says 'insufficient information' on sixty percent of its universe does not renew. A note that says 'high conviction' on everything renews twice. Clients read confidence as competence, and the feedback loop tightens every quarter. Nobody inside the machine is behaving irrationally. The machine itself is misaligned, and misaligned machines produce mispriced assets.
I learned the structural version of this at twenty-one, before I had capital to lose. In 2018, during the testnet migration season, I audited fifteen early token contracts. Off the shelf. No mandate, no payment, no client. Project Alpha shipped a standard token interface with a hand-modified transfer function. The community around it was loud, polished, and sincere. The deck promised a new financial primitive. The whitepaper was beautifully typeset.
I did not read the whitepaper for the audit. I read the bytecode. Inside the modified transfer function sat a classic integer overflow — the kind that lets you mint supply out of arithmetic rather than out of collateral. Estimated exposure at the time, roughly forty thousand dollars. I wrote it up. The founders rejected the report as 'too aggressive.' I published it on GitHub anyway. Three other security researchers cited it inside a year.
The lesson was never that I was right. The lesson was structural. The gap between the whitepaper and the bytecode is where every loss lives. And that gap is always filled by whoever is loudest, never by whoever is correct.
That is the null input problem stated plainly. When a field is empty, someone fills it. In a bull market, the filler is optimism. In a bear market, the filler is fear. Neither is a measurement. Both get logged as one.
There is a second, slower cost. Call it narrative debt. Every unverified claim that enters a memo and survives to a committee accumulates interest. Six months later it is cited as established fact in a follow-on memo by a different analyst who never saw the original source. The debt compounds silently until a liquidation event forces an audit, and at that point the entire chain of reasoning has to be reconstructed from memory. I have participated in three of those forensic exercises. They are always worse than the original loss, because they reveal that nobody can reconstruct the provenance of the position.
Core: A Null-Safe Framework
Nine dimensions is the correct number. Fewer and you miss a class of failure. More and the process becomes theater. The template I run is fixed, and the fixed part is not the questions. The fixed part is the rule for what happens when a question has no answer.
Rule One: Absence is a data point.
An empty field is not a neutral field. It is a measurement of the researcher's reach, and more importantly, a measurement of the project's transparency surface. A project with a live mainnet, a public repository, and a circulating token has a very large transparency surface. If nine dimensions of standard public information come back empty, that is not an information problem. That is a finding.
The absence of data is data. A project that cannot fill a standard template during a bull market is telling you something the deck will never say.
I keep a version of this rule for the Lightning Network, which I have tracked for seven years. The pitch is elegant: Bitcoin scaling, instant settlement, near-zero fees. The measured reality is a routing graph with persistent failure rates above a certain payment size, and channel liquidity management that is an operations job rather than a user experience. Seven years of tooling has not moved those failure rates to a place where a retail user completes a payment on the first attempt with high reliability.
The number that would settle the debate is first-attempt success rate on median payment size across the public routing graph. It is not prominently published, because the answer is not promotional. That is a null that has persisted across two complete market cycles. When a null persists across two cycles, stop treating it as a data gap. Treat it as the answer. Lightning is a settlement primitive for a narrow set of operators who are willing to run the operations. It is not, and will not become, a retail payment rail. The routing table has told us this every quarter; the narrative has overwritten it every quarter.
Rule Two: Audit the code, then audit the intent.
Provenance beats volume. A thousand posts are not one commit. A forty-page whitepaper is not one audited function. When I evaluate a claim, I rank sources in a strict order and I do not move up the order for convenience.
Tier one: deployed bytecode, verified against source, diffed against the previous deployment. Tier two: signed transactions on-chain, including deployment and admin calls. Tier three: repository activity with named contributors and merged pull requests. Tier four: filed governance proposals with executed outcomes. Tier five: documentation. Tier six: interviews, threads, and podcasts.
Tier six is where most conviction gets built. Tier one is where it should be built. A claim that lives only in tier six is a claim that has never been audited by anyone with the authority to break it.
There is a second half to this rule that took me longer to learn. The code tells you what is possible. The intent tells you what is likely. A forty-eight-hour timelock is a number. Whether the admin key sits with a multisig whose signers are all compensated by the same foundation is a different number, and it is more predictive of loss.
I apply this to the current layer-2 landscape, where the marketing fight is over proving systems. The recurring claim is that zero-knowledge rollups are technologically superior to optimistic rollups and will therefore win. That framing confuses the product with the distribution. The deployed reality is that the two stacks differ far less in current user-facing properties than in their go-to-market: which stack has convinced more teams to ship a chain under its umbrella. That is the variable that compounds. The winning rollup stack will be the one with the most teams deploying, not the one with the most elegant proof. Audit the code, then audit the intent — and when the code is roughly comparable, the intent is the entire story.
Rule Three: Nulls propagate. Model them.
In any dependency graph, a null in an upstream node contaminates every downstream node. This is not a metaphor. It is arithmetic.
Take token economics. If the allocation table is unknown, then circulating supply is unknown. If circulating supply is unknown, then fully diluted valuation is unknown. If FDV is unknown, then the valuation comparison against the competitive set is unavailable. If the comparison is unavailable, there is no basis for a position size. One null at the top of the chain zeroes out six fields at the bottom.
Most research processes handle this badly. They fill the top null with a reasonable-sounding assumption, then compute the entire downstream chain as though the assumption were an observation. The output looks rigorous. It has charts. It has a price target. It is fiction with a spreadsheet.
A derived number is only as auditable as its most speculative input. One assumption at the root makes the entire tree a hypothesis.
The correct behavior is to mark the whole branch unavailable and to size accordingly — which, in most cases, means not sizing at all. An unavailable valuation is not a neutral valuation. It is an unpriceable asset, and unpriceable assets belong on the watchlist, not in the book.
The unlock schedule deserves its own note. Almost every project discloses headline allocation percentages. Almost none presents the schedule in a form that lets you compute float in a specific quarter. That is not an oversight. The unlock schedule is the single most predictive document in a token launch and the single most frequently unavailable document. Where it is missing, assume the worst quarter is the one you cannot see.
Rule Four: Never divide by an unknown denominator.
This rule saves capital, and in a bull market it is violated most often.
Every ratio in finance has a denominator. Yield has a denominator: capital at risk. Market share has a denominator: total addressable volume. Adoption has a denominator: economically active addresses rather than addresses that received one dust transfer. When the denominator is unknown, the ratio is not approximately correct. It is undefined. And here is the part that matters. Undefined does not mean neutral; it means the maximum value will be quoted and the minimum will be realized.
I keep two clean case studies from my own book.
The first is from 2020. During the peak of the DeFi summer, I ran a personal portfolio of fifty thousand dollars across two lending and trading venues. Gas spiked to five hundred gwei. Most participants were making manual decisions and discovering that their risk parameters were a function of network congestion they had not modeled. I ran a pre-coded rebalancing script that unwound positions automatically when congestion crossed a threshold. It preserved ninety-two percent of capital while peers around me lost forty percent to slippage and failed transactions. I open-sourced the gas-aware library afterward, because the lesson was not proprietary. Efficiency beats speed, and a pre-coded rule beats a fast reaction every time congestion is the binding constraint.
The second is from 2021. During the peak of the profile-picture market, I held a floor position in blue-chip collections worth roughly one hundred twenty thousand dollars. The quoted floor was a number with a denominator nobody examined: the set of buyers willing to transact at that floor. When demand thinned, the denominator collapsed before the numerator did. I ran a pre-committed stop at fifteen percent drawdown and sold sixty percent of the position inside one hour, preserving roughly seventy thousand dollars of liquidity. Peers held and waited for a rebound. There was no rebound at that floor, and the reason was arithmetic, not sentiment. The floor had never been a measurement of value. It had been a measurement of queue depth on the bid.
Ledger books, not feelings, settle the debt.
The same denominator problem sits underneath the entire interoperability thesis, and I will return to it. Hold the thought.
Rule Five: Install circuit breakers, and wire them before the trade.
A circuit breaker is a pre-coded halting condition. It is not a judgment call. If it were a judgment call, it would fail, because judgment degrades precisely when it is needed most.
In 2022 I was running oversight for a small trading desk when an algorithmic stablecoin broke its peg mechanics from the inside. Before that week, I had mandated a halting condition across all algorithmic stablecoin strategies: if the peg deviates beyond a defined band and the deviation does not correct within a defined window, every related strategy stops. The desk halted roughly thirty seconds ahead of the main cascade. Competitors running judgment-based oversight did not stop. They were still debating whether it was a buying opportunity while the liquidity was already gone.
Liquidity dries up when confidence breaks. That is not a slogan. It is the mechanism. The order book does not thin gradually; it evaporates, because market makers are not price-discovery agents. They are inventory-risk agents, and when inventory risk becomes unpriceable, they remove quotes rather than widen them. Widening implies a price at which you would transact. Removal implies there is no such price.
After that week I standardized position limits across every asset the desk touched. The reputation shift inside the firm was more instructive than the P&L. The operator who halts is not seen as cautious. The operator who halts before a cascade is seen as reliable. Institutional trust is not built on returns. It is built on the predictability of when you stop.
The most recent version of this is the 2025 reporting standard I built for an institutional options mandate: a five-million-dollar delta-neutral structure using Ethereum call spreads, reported on Vega and Theta exposure only. Directional commentary was removed from the template entirely because it introduced noise into a book that was designed to be directionless. Removing the noise let the client execute without second-guessing, and the structure returned fifteen percent risk-adjusted across a volatile quarter. A template that shows only the variables you are actually trading is not a simplification. It is a control.
Null Handling Across the Nine Dimensions
Apply the five rules and the template becomes a decision system rather than a checklist.
Technical. Innovation, maturity, security assumptions, performance. The usual null is maturity: concept, testnet, or mainnet with production traffic? If the answer is unavailable, the entire dimension is unavailable regardless of how elegant the concept is. A concept is not a protocol.
Token economics. Type, supply model, allocation, unlock schedule, incentive sustainability, value capture. The critical null is the schedule, and the scheduling risk is asymmetric. Incentives paid in a native token are sustainable only if real revenue backs them. In most designs, it does not, and the structure reduces to new capital paying old capital. That is not a governance opinion. It is a cash-flow statement.
Market. Event type, degree of pricing-in, expected volatility, regime. The usual null is 'how much of this is already in the price,' and the honest answer is that nobody knows. The disciplined response is to express the view in a structure that does not depend on that answer. That is what an options desk is for, and it is why I moved to one.
Ecosystem. Position in the supply chain, upstream dependencies, downstream integrators, developer signal, user signal. The usual null is retention. Daily actives are cheap. Economically active retention is expensive to fake and almost never disclosed. Default assumption: if retention is not presented, it is not good.
Regulatory. Facts-and-circumstances securities analysis, jurisdictional exposure, sanctions surface, KYC posture. The usual null is the entity structure, and in 2026 that null is unacceptable. A project selling tokens into a major market without a disclosed legal structure has priced regulatory risk at zero. Zero is never the right price for a non-zero probability.
Team and governance. Named or pseudonymous, historical execution, prior failures, governance participation, voter concentration, investor quality, lockups. The usual null is failure history. Every founder has a history. Where a verifiable history is absent, assume the absence is the history.
Risk. Technical, market, operational, regulatory, competitive, narrative. The usual null is the severity pairing. Risk without probability and impact is decoration. A matrix full of N/A is not a clean bill of health. It is a statement that nobody has done the work.
Competitive set. Who else is doing this, with what liquidity, at what market share. The usual null is genuine differentiation, and in a bull market it is almost always absent.
Provenance. Where did each claim originate, and at which tier. This is the null that should stop the process. If the provenance chain cannot be reconstructed, no claim in the document is trustworthy, and the entire note is a tier-six artifact.
Contrarian: More Coverage Is Not More Certainty
The standard response to the null input problem is to demand more research. More dimensions, more feeds, more dashboards, more analyst hours. This is the wrong cure, and it is actively harmful, for three reasons.
First, coverage is not verification. Depth of coverage is inversely correlated with depth of verification, because attention is a fixed budget. A team tracking four hundred assets verifies none of them. A team tracking twelve can read the bytecode on all twelve. The number of fields in your template is not a measure of rigor.
Second, a larger template manufactures more nulls, and more nulls require more filling. Every added dimension is another opportunity to interpolate, and interpolation is where errors enter the system. This is why my template is fixed at nine and why I will not expand it for a single impressive deal. The constraint is the control.
Third, the industry rewards the appearance of coverage. A research product that reports insufficient information on sixty percent of its universe does not sell. A product that reports high conviction on everything sells, because allocators read confidence as competence. The market pays for certainty. It does not pay for correctness, at least not on the timeline that determines the analyst's bonus.
In a bull market, demand for certainty is elastic and the supply of certainty is fabricated. That is the entire failure mode of the research industry, and no volume of additional data fixes it.
Now the interoperability thread, because it is the same error at protocol level.
The pitch for cross-chain infrastructure is that more connections equal more liquidity. Fifteen bridges. Twenty. Forty. Every new route is described as expanding access. The measured effect is the opposite. Every additional chain fragments the liquidity that already exists, because market makers must now inventory the same asset across more venues, and inventory across more venues means less depth per venue. Depth per venue determines slippage. Slippage is what the user actually experiences.
More interoperability is not more liquidity. It is the same liquidity, sliced thinner, across more points of failure.
The empirical record is not ambiguous. The largest losses in this market's history did not come from a chain failing. They came from a bridge failing. The bridge is the most concentrated point of trust in a system that markets itself as trustless. Every new bridge is a new concentrated point, and every new point is a new audit surface. Audit surfaces fail at a rate roughly proportional to their number.
This is not an argument against interoperability as a concept. It is an argument against the accounting. Count the venues, then count the depth per venue. If depth per venue falls while the number of venues rises, you have not solved the liquidity problem. You have renamed it.
The same arithmetic sits under the layer-2 conversation. Every additional rollup is described as scaling. What it actually does is slice activity across more sequencers, more bridges, more liquidity pools, and more fragmented state. The teams that understand this are not competing on proving systems. They are competing on getting the most projects to deploy under their umbrella, because distribution compounds and liquidity does not. That is the real question between the two major stacks, and it has almost nothing to do with the proof.
Takeaway
Back to the March memo, and to the frame that actually matters.
The correct output for that intake was not a question. It was a status: no basis for valuation, no basis for sizing, watchlist, revisit on disclosure. The project may be excellent. It may return forty times. That is not information that was available, and acting as if it were available is not conviction. It is leverage against an assumption.
What belongs on the wall of every research desk:
- A null is a finding, not a gap. Write it down and act on it.
- Rank sources. Never promote tier six to tier one for emotional convenience.
- Propagate nulls through the model. Never fill the root with a reasonable-sounding number.
- Never divide by an unknown denominator. Undefined ratios get quoted at maximum and realized at minimum.
- Wire circuit breakers before the trade, not during the cascade.
- Fixed templates, not expanded ones. The constraint is the control.
The forward-looking question is not whether the next cycle produces another round of projects that cannot fill a standard template. It will. The question is whether the research function inside your firm has the structural authority to return 'insufficient information' on a deal that four respected funds have already funded — and whether the person who types that string is rewarded or replaced.
Bull markets manufacture certainty and bill it to the next cycle. The firms that survive are not the ones with the largest coverage universe. They are the ones whose templates halt instead of guessing, whose analysts are paid for refusal as readily as for conviction, and whose ledgers contain only entries that someone, somewhere, verified against deployed bytecode.
Ledger books, not feelings, settle the debt. The same is true of research: only the verified entry survives settlement. Everything else is a rounded number on a slide, waiting for the denominator to arrive.