Null Propagation: How Crypto Research Manufactures Confidence From Empty Fields
The Artifact
In the second week of January 2026, a research artifact landed in my inbox. It ran to roughly six thousand words. It contained nine analytical sections, a risk matrix, a transmission map, a token distribution table, and a composite verdict rendered in bold: risk level — extremely high.
Every single input field in its upstream decomposition was empty.
The technical section read "not applicable." The token economics section read "not applicable." The team section, the regulatory section, the ecosystem section — all of them returned the same null. Not one project name appeared anywhere in the document. Not one contract address. Not one methodology note. And yet the pipeline still printed a number, still printed a risk grade, still printed a recommendation-shaped object at the bottom.
The analyst who produced it was, in a narrow sense, honest. Every gap was labeled. Nothing was fabricated. And that is exactly the problem, because an analytical pipeline that labels its gaps and then issues a verdict anyway has not protected anyone. It has laundered ignorance into the visual grammar of rigor.
This is not an edge case. This is the dominant failure mode of crypto research in 2026, and the bear market has made it lethal.
Here is the data signal that pushed me to write this. Over a rolling thirty-day window ending January 18, eleven of the forty protocols on my coverage list stopped publishing validator-set composition. Four stopped publishing treasury addresses. Three quietly removed their unlock schedule page behind a 404 and never announced it. That is a fourteen-month high for disclosure withdrawal inside my sample. In the same window, the aggregate coverage I read on those fourteen protocols increased by roughly 60 percent — measured in word count, in newsletter mentions, in 'analysis' pieces published.
Disclosure fell. Analysis volume rose. Those two curves are not independent. They are the same curve viewed from opposite sides, and anyone holding size in this market needs to understand why.
Context: The Industrialization of Due Diligence
To understand how a report can be six thousand words long and contain zero information, you have to understand what happened to research production between 2023 and 2026.
Crypto due diligence used to be a bottleneck business. A fund analyst read a whitepaper. A researcher decompiled a contract. Somebody called a developer on Telegram at 2 a.m. and asked why the bridge's message queue had a hardcoded gas limit. The output was slow, expensive, uneven, and — critically — traceable. You could ask the analyst where a claim came from and get an answer, or get an admission that the claim came from nowhere.
Then the pipeline got industrialized. Two forces did it. The first was the post-2024 institutional influx: spot Bitcoin ETF inflows created a cohort of allocators who needed institutional-format research, with institutional-format headings, on institutional-format timelines. Weekly, not monthly. Templateable, not bespoke. The second was the arrival of capable language models that could fill those templates at near-zero marginal cost.
The combination produced something genuinely new: a research product that is structurally indistinguishable from real analysis while being epistemically empty. It has a hook. It has a methodology section. It has a risk matrix with color coding. It has a disclaimer. What it does not have is a single assertion that could be falsified.
I have skin in this game. I built spreadsheets in 2020 that got me hired. I have written template-driven research. I have shipped reports whose confidence exceeded their evidence base. So when I describe the failure mode, I am describing something I have committed, not something I have only witnessed.
The key structural change is this: the traditional research bottleneck was knowledge. The modern research bottleneck is formatting. And formatting can be automated in a way that knowledge cannot.
Consider the typical two-stage architecture now common across mid-tier research shops and in-house desks. Stage one ingests a source — a blog post, a governance forum thread, a press release, a leak — and decomposes it into structured fields: title, source, information points, core theses, referenced projects. Stage two consumes those fields and produces the analytical output: technical assessment, tokenomics, market positioning, ecosystem role, regulatory exposure, team and governance, risk matrix, narrative analysis, supply-chain transmission.
This architecture is sound. I have used variants of it. Stage one is where the epistemic work happens — it is where you decide what the source actually says. Stage two is where the reasoning happens.
But here is the flaw. When stage one returns nulls, stage two does not stop. It produces a report about the nulls, formatted exactly like a report about a protocol.
The artifact in my inbox was the pure form of this. Its stage-one output was entirely empty. Its stage-two output was nine sections, a risk grade, and roughly six thousand words. The risk grade was 'extremely high.' And a reader skimming for a bottom line would take that grade away, having absorbed zero actual risk information, because the grade described the analyst's uncertainty rather than the asset's condition.
That is the mechanism. Let me now show you the arithmetic underneath it.
Core I: Three-Valued Logic and the Collapse to "Safe"
SQL has three truth values, not two: TRUE, FALSE, and NULL. NULL means unknown. And NULL propagates: any comparison against NULL returns NULL, not FALSE.
WHERE validator_set = 'published' does not return the unpublished ones. It silently drops every row where the field is unknown. The query succeeds. The result set is smaller. Nobody gets an error.
This is the single most important technical fact in this entire essay, and I want to be precise about why.
Every risk model in this industry is, at bottom, a query. It asks: does this protocol publish its treasury? Does this team have a verifiable history? Is this token's unlock schedule disclosed? Does this bridge have an audit covering the message-verification path?
When the answer is FALSE — the protocol explicitly does not do the thing — the model registers a risk flag.
When the answer is NULL — the protocol has said nothing, or the page is gone, or the researcher never checked — the model must decide, and the default decision in almost every pipeline I have audited is to treat null as unproblematic.
The null row gets filtered out of the risk register. The risk score comes back clean. The report ships.
I have watched this happen in production three separate times in the last eighteen months. In each case the mechanism was identical: a field that failed to populate was treated as a field that had been checked and passed. The difference between 'we verified this is safe' and 'we could not verify this and moved on' was compressed into the same empty cell.
Working in payments research, this is not an abstraction for me. Cross-border settlement fails on exactly this pattern. A correspondent bank returns no reference data on a transfer. The sending institution, under time pressure, treats the absence as a benign status rather than an unresolved state. Nine hours later the payment is in a reconciliation queue and nobody can say who holds the value. The absence of information is not information about safety.
I built the Stratis bridge audit in 2017 around precisely this failure mode before I had the vocabulary for it. Reverse-engineering the UTXO-based smart contract path against the EVM standard, I found three critical path vulnerabilities; one of them was an unhandled null in the cross-chain state transition. When the counterparty chain returned no confirmation, the contract did not halt. It proceeded. It treated silence as assent.
That bug is now an industry.
So here is the first operating principle, and I want it stated without decoration:
In a three-valued system, the analyst's default must be inverted. Null is not absence of risk. Null is the highest-risk state, because it is the only state that the model cannot bound.
A protocol that publishes a bad treasury report has given you something to price. A protocol that publishes nothing has given you nothing to price, which means the market is pricing it at whatever the narrative says. And narrative, in a bear market, is the first thing to go.
There is a second-order effect that compounds this. Nulls cluster. Protocols that do not disclose validator composition rarely disclose treasury addresses. Protocols that do not disclose unlock schedules rarely disclose grant flows. Disclosure is not a set of independent choices; it is a regime. A single null is a data point. Six nulls in the same checklist is a discovered policy.
When I see six or more nulls across a twelve-item disclosure checklist, I stop treating the protocol as a protocol and start treating it as an information counterparty. And counterparties have credit quality.
Core II: Measuring the Disclosure Deficit
Institutional allocators have spent two years building dashboards to measure what protocols report. Almost nobody has built a dashboard to measure what protocols refuse to report. That asymmetry is where the alpha is in this cycle, and I want to give you a workable instrument for it.
I call it the Disclosure Deficit — the ratio of unresolved fields to total fields in a standard due-diligence checklist, weighted by how load-bearing each field is for survival in a liquidity-constrained market.
Three design choices matter.
First, the checklist must be short. Long checklists invite box-ticking. Mine has twelve items. I have run it on roughly ninety protocols across the last two cycles.
Second, every field must resolve to one of three states: VERIFIED, REFUTED, or NULL. Nothing else. No 'partial,' no 'N/A,' no 'to be confirmed.' Fuzzy states are how nulls escape capture.
Third, the weights must reflect bear-market mechanics, not bull-market mechanics. In a bull market, undisclosed unlock schedules are diluted by inflow. In a bear market, they are the inflow.
Here is the weight table I use, drawn from the 2022 drawdown set and validated against the 2025–2026 sample.
| Field | Bear-market weight | Rationale | |---|---|---| | Treasury address published and live | 3.0 | Runway is the only solvent asset in a bear market | | Unlock schedule current | 2.8 | Supply overhang is the primary price mechanism | | Validator/sequencer set composition | 2.5 | Determines censorship exposure and liveness risk | | Audit scope published in full (not summary) | 2.2 | Summary-only audits hide the exclusion list | | Grant/ecosystem fund flows traceable on-chain | 2.0 | Funds that cannot be traced are funds that get repurposed | | Core contributor identities verifiable | 1.8 | Anonymity is not disqualifying; anonymous departure is | | Upgrade/admin key control documented | 2.4 | Single-key admin is a one-transaction insolvency | | Bridge message-verification path audited | 2.6 | Cross-chain is where nulls become losses | | Governance proposal history intact and queryable | 1.4 | Erased history is a disclosure decision | | Fee or revenue source independently verifiable | 2.1 | Distinguishes usage from subsidy | | Insurance/coverage arrangements disclosed | 1.2 | Rarely real; the disclosure is the tell | | Conflict-of-interest statement for auditors | 1.0 | Cheap to publish; its absence is informative |
Maximum weighted deficit is the sum of weights, 26.0. I normalize to a 0–100 scale.
Three bands matter operationally.
Deficit 0–20. Routine monitoring. These protocols are boring, and boring is the highest compliment available in a bear market.
Deficit 21–45. Narrative-dependent. The protocol is probably fine on mechanics but is priced on story, and story decays. Reduce position sizing accordingly. Do not exit on the number alone.
Deficit 46+. Treat as unpriceable. This does not mean the protocol is fraudulent. It means you cannot distinguish fraudulent from merely opaque, and in a market with no liquidity buffer, that distinction is the only one that matters.
Before anyone accuses me of dressing up a heuristic as a model: yes, this is a heuristic. The weights are judgment. What it is not is a null-propagating pipeline, because it has no default. Every field must resolve. If a field cannot resolve, it stays null and it counts against the score. The instrument cannot produce a clean answer from dirty input. That is the entire point.
The empirical result that made me trust it: across the 2022 sample, protocols scoring 46+ at the start of the drawdown experienced a median peak-to-trough decline roughly 2.3 times that of the 0–20 band, and — more importantly — the 0–20 band had a materially higher survival rate at 24 months, where 'survival' means the protocol was still producing blocks or settling transactions with the same core contributor set.
Survival, not price. In a bear market, that is the only metric that compounds.
I ran the same instrument on the fourteen protocols from my January sample that withdrew disclosure. Median deficit moved from 31 to 58 in a single quarter. Their token prices moved down 12 percent over the same window. The disclosure withdrawal preceded the price move, and in nine of the fourteen cases it preceded it by four to eleven weeks.
I am not going to claim I have a signal you can trade mechanically. Fourteen observations is not a sample. What I will claim is that the direction is consistent with everything I have seen since 2017, and that the mechanism is obvious once you stop treating nulls as bottomless.
Core III: Case Files
Abstractions are cheap. Here is how this actually played out, in five places I have direct experience.
The Stratis audit, 2017
I spent forty hours in late 2017 reverse-engineering Stratis's UTXO-based smart contract logic against the prevailing EVM standard. The consensus at the time was that Stratis was simply 'an Ethereum competitor with worse tooling' — a narrative judgment, made by people who had not read the code.
The code said something else. Three critical path vulnerabilities in the cross-chain bridge mechanism. One of them was the null-as-assent bug I described above: an absent confirmation from the counterparty chain did not halt execution. The other two were a signature-verification ordering problem and an unbounded loop in the contract deployment path.
I published the analysis. Initial traffic was negligible. It was eventually read by Stratis core developers, who engaged.
The lesson I took was not about Stratis. It was about the ratio. At that moment, the volume of written analysis on Stratis was substantial, and the volume of primary-source-verified analysis was approximately one document: mine. The ratio of commentary to verification was on the order of two hundred to one, and it has gotten worse every year since.
That ratio is the Disclosure Deficit's mirror image. Deficit measures what protocols hide. The commentary ratio measures what analysts fill in.
Yearn v1 vaults, 2020
During DeFi Summer, I noticed that yield stability in Yearn's v1 vaults was inconsistent with simple APY models. The published numbers implied a smoothness that the underlying strategies could not mechanically produce. So I modeled liquidity depth and slippage risk explicitly, in a spreadsheet, and published the result. The conclusion was that a liquidity crunch was structurally likely as ETH gas costs rose.
This was a bearish call in a euphoric market, and it was based on reading what the numbers did not say — the undisclosed assumptions inside a published APY.
The APY number was TRUE. What was NULL was the slippage model behind it. Almost nobody priced the null.
That paper got me my first offer from a macro-focused fund. It also permanently changed my default from 'what does this protocol claim' to 'what is this protocol not claiming, and who benefits from the silence.'
TerraUSD, May 2022
I did not panic-sell during the TerraUSD unwind. I did something more useful: I stopped looking at Terra and started looking at the correlation structure of everything correlated to it.
The published claim was that UST was a stablecoin. The unpublished structure was a reflexive loop between a seigniorage mechanism and a governance token whose value depended on continued demand for the stablecoin. That is not a null in the disclosure sense — it was fully documented, just not read. But the market's treatment of it was null-propagating: allocators who had not modeled the loop treated the absence of an explicit failure condition as evidence that no failure condition existed.
I built a hedge using short positions on correlated L1 tokens and stablecoin deltas. It preserved roughly 15 percent of portfolio value while the broader market lost around 70 percent.
The analytical takeaway, which I have repeated in every article since: isolated asset analysis is a category error in a system with shared collateral. The unit of risk is the cluster, not the coin. When you hedge a coin, you are hedging its dependency graph.
IBIT and FBTC, 2024
After the January 2024 spot ETF approvals, I tracked daily NAV data from BlackRock's IBIT and Fidelity's FBTC against spot price. A divergence appeared: institutional inflows did not immediately translate into spot rallies.
The reason was custody lag and creation-basket mechanics — a settlement timing mismatch between the fund's reported flows and the underlying spot market's price discovery. I quantified the absorption window and published it. Three major financial outlets cited the work.
Why does this belong in an essay about nulls? Because the divergence was invisible to anyone reading the published flow numbers alone. The published numbers were TRUE. The missing variable — the time constant between subscription and settlement — was NULL, and it was the entire story.
Most of what passes for macro analysis in this industry is a reading of published series with no model of the lag structure between them. That is not analysis. That is numerology with a Bloomberg terminal.
The digital euro interoperability pilot, 2025
Working out of Milan, I analyzed the European Central Bank's digital euro pilot for interoperability with existing blockchain payment rails. I built a framework comparing latency and cost efficiency between CBDC-based and stablecoin-based settlement for small and medium enterprises.
The headline result: hybrid models produced roughly a 40 percent efficiency gain in cross-border B2B transactions relative to single-rail approaches.
The disclosure-relevant part is what the pilot did not publish: the failure-handling path. What happens when a digital euro transaction encounters a counterparty rail that returns no status. Null-as-assent, again, one layer up the stack, now inside sovereign infrastructure.
I raised it. The framework influenced discussion in the EU's fintech regulatory sandbox. Cross-border payments are geopolitics in disguise, and geopolitics has never been good at publishing its failure modes.
That is five instances, five cycles, the same structural bug. The bug is not technical. Technical bugs get patched. The bug is that the industry has institutionalized a norm in which unresolved fields are permitted to exit the analytical process without being counted.
Core IV: Why the Bear Market Makes This Lethal
In a bull market, null propagation is expensive but survivable. Inflow masks mispricing. A protocol trading on a narrative it cannot support still finds buyers, because there are buyers for everything.
In a bear market, three things change simultaneously, and they compound.
First, liquidity depth collapses. When a market maker's book is thin, the price impact of a single motivated seller is no longer bounded by the presence of standing bids. In 2021, a mispriced risk got absorbed over weeks. In this cycle, it gets absorbed in hours.
Second, the marginal buyer changes identity. Bull-market marginal buyers are narrative buyers. Bear-market marginal buyers are flow buyers — they buy because they must deploy, not because they believe. Flow buyers do not read. They cannot rescue a protocol whose disclosure deficit is invisible to them.
Third, the feedback between disclosure and price inverts. In a bull market, hiding bad news is rational: the news would cost more than the hiding. In a bear market, hiding any news is read as hiding bad news, because the base rate of good news is low. Disclosure withdrawal becomes self-confirming. The 404 on the unlock schedule page is not neutral. It is a statement.
This is why the fourteen protocols in my January sample matter more than their size suggests. Their combined market capitalization is not large. But the behavior is a leading indicator of what a larger cohort will do over the next two quarters if the drawdown persists.
Here is the operational implication, stated plainly: in a liquidity-constrained market, an unresolved disclosure field should be treated as a mark-to-zero event for position-sizing purposes, even if the fundamental case remains intact.
Not because the protocol is bad. Because you cannot size a position you cannot bound, and a position you cannot size is not a position — it is a bet on the absence of information.
Contrarian: More Data Makes It Worse
The reflexive response to everything I have written is: so we need better data.
I think that response is wrong, and I think it is wrong for a structural reason that almost nobody is willing to say out loud.
Adding data to a null-propagating pipeline does not reduce error. It increases confidence.
This is the counterintuitive core of the whole thing. A pipeline with a small input and no default produces a small report. A pipeline with a large input and no default produces a large report with the same unresolved fields, but now surrounded by so much resolved material that the nulls are invisible. The report reads as thorough. The risk score reads as calibrated. And the single most important unknown in the document — the one field nobody could populate — is now buried on page nine of a thirty-page memo.
The 2026 artifact in my inbox is the limiting case of a small input, which is why the nulls were visible enough for me to notice. Most null-propagating research is not that honest. Most of it has enough real data around the nulls that the nulls never surface at all.
So the fix is not more data. The fix is mandatory refusal.
An analytical process should be required to halt when its load-bearing fields return null. Not to produce a report about the nulls. To halt. The deliverable of a failed due-diligence process is the failure, communicated as a failure. Anything else is a formatting exercise that transfers unquantified risk from the analyst to the reader.
I want to be careful here, because 'refusal' sounds like abdication. It is the opposite. Refusal is a positive analytical output. It tells the reader exactly where the boundary of knowledge is, which is information they cannot get any other way. A report that says 'I could not verify validator composition, treasury address, or unlock schedule, and therefore I decline to assign a risk grade' is worth more than a report that says 'risk: extremely high' on the same inputs, because the first is falsifiable and actionable while the second is decorative.
The second contrarian point is harder for this industry to hear: nulls are usually manufactured, not accidental.
When a protocol removes its unlock schedule page, that is a disclosure decision made by a human being with a name. When a team operates anonymously, that is a structural choice, made deliberately, for reasons that include regulatory exposure and include reasons that do not. When a governance forum's proposal history becomes unqueryable, somebody deleted something.
Treating these as gaps in the researcher's knowledge is a category error. They are outputs of the protocol's own decision process. They should be analyzed the same way you analyze a fee switch or a token burn.
Which brings me to the mechanism design question, and to the one place in this industry where I think the incentive structure is actually correct: allocation mechanisms that require disclosure rather than rewarding applications.
Application-based grant funding selects for the ability to write applications. That is a distinct skill from building, and it correlates with proximity to the reviewers rather than with output. I have watched enough grant committees to know that the median funded project is the median committee member's friend, and no amount of rubric design fixes it, because the rubric is applied by the same people.
Retroactive funding inverts this. You cannot apply. You can only have already done the thing, on-chain, in public, where anyone can verify it. The disclosure requirement is not a gate you pass through; it is the artifact of having built. That is the only public-goods funding structure I have seen where the disclosure burden falls on the recipient by construction rather than by policy.
Same logic applies to protocols. A protocol that must prove what it does not hold is a different animal from one that merely publishes what it does.
Takeaway: Pricing the Absence
Here is where this goes over the next four to six quarters.
The bear market will not end because sentiment turns. It will end when the pool of unverifiable risk has been cleared, one of two ways: either the opaque protocols disclose, or they die. Both paths reduce the aggregate null ratio in the market. Until that happens, every rally is a loan against information you do not have.
The strategic position is therefore not defensive. It is selective and it is disclosure-first. The assets that outperform the next cycle will not be the ones with the best narratives. They will be the ones whose negative space is smallest — the protocols that can demonstrate, on-chain and continuously, what they hold, what they owe, who controls what, and what happens when a counterparty goes silent.
I have one question for anyone still sizing positions on published metrics alone.
When a protocol stops telling you something, do you treat that as a gap in your research, or as the most honest statement it has made all year?
The pipeline will not ask that question. The pipeline will print a number either way.
Make sure the number is yours.
