The number is impossible to ignore: 62.3% versus 21.01%. That is not a rounding error. It is not a market blip. It is the positive-outcome rate on WSOL balance changes for twelve addresses that habitually routed through a proprietary AMM called HumidiFi, measured against every other trade in their own peer group over the same observation window. The ratio is 2.97. The venue is Solana. The data was public all along. Nobody connected the dots until an academic paper did.
Let me be precise about what bothers me. Every actor in this sample shared the same chain. Same RPC layer. Same Jupiter aggregation surface. Same mempool-level transparency Solana's design affords. Yet twelve addresses captured a systematic three-times profitability edge because they possessed a single piece of non-public knowledge: a closed liquidity venue exists, and routing through it pays.
The code whispered secrets the audit missed. That is not a bug report in the classical sense. No reentrancy. No integer overflow. No governance backdoor. The vulnerability is structural. A closed execution venue embedded in an open network produces information asymmetry that no fee schedule, no gas mechanism, and no priority-fee auction can fully arbitrage away. Eleven years of auditing crypto systems have taught me one durable lesson: when profitability gaps of this magnitude appear between two groups operating at identical infrastructure, look at what they know, not what they do.
The Study Behind the Number
The research is not a blog post or a Twitter thread. It is a peer-context empirical study accepted at ASE 2026, a software engineering venue with real methodological standards. The authors sampled 200 Solana addresses classified as MEV-like, isolated a one-month observation window covering October to November 2025, and reconstructed every transaction in those addresses' histories through Solscan program-ID annotations. 463,411 non-noise trades were identified. 244,733 of them touched HumidiFi.
The study went further than most academic work in this space. The researchers analyzed 586 open-source bot codebases to understand the broader automation ecosystem. They labeled two commercial bot services, Trojan and SolanaMevBot, and identified structural divergence between groups of addresses based on behavior. They published a replication package on Zenodo. This is not a headline with a lazy summary; this is a reproducible forensic slice of a live market.
The central finding is stark. A twelve-address cluster exhibited a 62.3% positive-outcome probability on WSOL balance changes when transactions involved HumidiFi, against 21.01% for their other trades. The authors were appropriately conservative: they flagged the gap as associated with HumidiFi access, not necessarily caused by it. They noted that routing "may involve closed or specialized liquidity," a hedge that quietly encodes the paper's most damning implication.
WSOL appears in between 70.90% and 99.94% of transactions across the MEV-like clusters. That makes Wrapped SOL the settlement currency of Solana's shadow economy, exactly as WETH anchors Ethereum's MEV complex. The paper also observes a separately concentrated cluster of 102 addresses with 80.9% of their activity inside the Pump.fun ecosystem, a meme-coin cohort with a distinct behavioral fingerprint. The twelve-address cluster is not the largest group and not the most active overall. It is simply the most profitable one.
That raises the obvious question. If the data is openly indexed, if the transactions are public, if Solscan can annotate the program IDs, why has no one identified HumidiFi and flooded it with copycat volume? The answer sits in the gap between data availability and attention. The chain publishes everything. It takes a skilled reader to notice what the chain publishes quietly.
Anatomy of an Information Edge
Let me describe what a proprietary AMM actually does inside Solana. Public venues like Raydium and Orca broadcast their liquidity depth through standard interfaces. Jupiter aggregates those venues, and its routing engine discovers arbitrage opportunities across the ecosystem. Consequently, any price deviation that appears on a public AMM is visible to every search bot within seconds and is generally eliminated within a block or two. Competition erodes the edge. That is the efficient-market story Solana's supporters tell.
HumidiFi breaks the story. If HumidiFi does not broadcast its full liquidity state to public aggregators; if it sits outside Jupiter's default route-candidate set; then deviations inside its pools are visible only to machines that know the program ID and actively query it. A twelve-address cluster knew. That single fact generates three compounding advantages.
First, discovery lag. Because public search bots do not watch HumidiFi's pools as closely as they watch Raydium, pricing errors persist longer. The cluster can capture deviations that a public-venue arbitrageur would erase within seconds. Slower competition means wider captured spreads. The 62.3% positive-outcome rate is precisely what an edge-lag profile looks like before competitive forces compress it.
Second, failure-rate reduction. Solana's public MEV competition is brutal. Priority-fee auctions, contested blocks, and constant arbitrage churn produce high drop rates. A closed venue with a smaller set of competing callers means fewer transactions are invalidated by state conflicts or by being outbid. Lower drop rates matter enormously for outcome statistics. A bot that successfully lands 80% of its transactions has a fundamentally different profitability curve than one landing 50%.
Third, sandwich protection. On a public AMM, an arbitrage transaction itself becomes a target for other MEV bots. The moment a bot detects an incoming profitable trade, it can wrap that trade in a front-run and back-run, extracting the value that the original trader expected to capture. A closed venue dramatically reduces this vector. The cluster's transactions are not being replayed, front-run, or sandwiched at the same rate as public-market trades, which inflates the conditional profitability of HumidiFi routing.
There is also a more subtle mechanism: the pool itself may price differently. If HumidiFi's smart contract implements a non-standard curve, includes no fee-sharing hook, or applies maker-taker logic that public AMMs do not, its internal prices can drift away from cross-venue equilibrium. The cluster that understands those pricing mechanics is effectively running a private microstructure strategy in public.
All of this remains inference. The paper does not decompile HumidiFi's contract. It does not publish the program bytecode. It does not simulate the pool's pricing logic. What it shows is behavioral evidence consistent with each of these mechanisms. For an auditor, that distinction is everything. I observe the effect; I cannot verify the mechanism. Between the lines of bytecode lies the trap; and I cannot read the bytecode because the owners have not chosen to expose it.
The honest summary: HumidiFi's advantage is real in its observed outcomes and unknown in its origins. That is the precise definition of a black box. The market has priced it as noise. The study says it is signal. Both can be true only if the signal is decaying, but the observation window says otherwise. Over 244,733 calls in one month, the cluster kept routing through HumidiFi. Either the owners were systematically fooling themselves, or the edge was durable enough to survive a month of continuous exploitation.
The Black Box Problem
Let me apply the standards I use in my own audits. When I review a protocol, the first question is not "does it make money" but "can I verify why." Public AMMs provide source code, bytecode verification, formal specification audits, and transparent liquidity. Anyone can simulate their behavior, measure their depth, and compare realized prices against a cross-venue index. HumidiFi provides none of that.
There is no public audit report. There is no open-source repository. There is no team disclosure. There is no token. There is no governance forum. There is no documentation describing the curve, the fee structure, or the router logic. The only evidence of its existence is the transaction trail. In any other regulated market context, this protocol would be classified as an anonymous trading venue and shut down for failing transparency requirements. On Solana, it simply exists, anonymous, unbranded, quietly executing billions of dollars of value.
I spent six weeks reverse-engineering the UST depeg mechanism in 2022. I traced token flows, isolated the mint-burn feedback loop, and concluded that the mechanics made bankruptcy mathematically inevitable months before the market agreed. The methodology worked because the code was visible. Terra's contracts were public. The flaw was not hidden; it was ignored. HumidiFi inverts that situation. The edge is hidden. The effect is visible. And because the contract is invisible, the same data cannot be subjected to first-principles verification.
Without bytecode-level analysis, I cannot conclude whether HumidiFi's 62.3% positive-outcome rate comes from a fundamentally better pricing function, a deliberately exploitative fee structure, a privileged data-provider relationship, or pure latency edge. Neither can the researchers. This is not a failure of their methodology; it is a structural property of closed execution venues. And it carries direct risk for any actor attempting to replicate the cluster's success. A trader who blindly routes funds to HumidiFi based on a behavioral study is trading on a hypothesis, not a verified mechanism.
I do not trust; I verify the hash. The hash is unavailable. Therefore, the rational conclusion is not "HumidiFi is profitable" but "HumidiFi's profitability is unverifiable, and its sustainability is unknown." In my audit reports, that distinction is not a nuance. It is the entire finding.
There is a deeper concern. If HumidiFi's advantage is real and structural, its owners sit on a commercial goldmine. If the same entity controls both the AMM and the bot cluster, then the observed "routing advantage" is actually internalization: the crypto analog of a proprietary trading desk executing against its own matching engine. The study cannot rule this out. The distinction matters because an internalized pool has no incentive to publish its fee schedule, no incentive to provide fair prices, and no external pressure to improve. It becomes a self-contained extractor of cross-venue inefficiency, appearing to the outside world as a favorable routing destination while functioning as a private toll road.
I encountered a milder version of this in a modular blockchain audit earlier this year. The sequencer selection algorithm concentrated block production in a small set of addresses, effectively creating a latent censorship surface. The team insisted on shipping. I insisted on redesigning. The delay cost two months and prevented a potential freeze of roughly fifty million dollars. The principle is identical: closed operational constraints inside an open network always produce extractive asymmetries.
Correlation, Causation, and the Survivorship Trap
The paper is rigorous about its own limits. It labels the 62.3%-versus-21.01% result as association. It notes that the profitability metric, WSOL balance changes, has not been stripped of fees, tips, timing, or address behavior. The authors acknowledge that the cluster differs in unobserved ways from its peers. This is academic punctiliousness. But it is also a warning. The reader who universalizes the finding, who believes that calling HumidiFi's contract from any address will generate a three-times outcome advantage, is misreading the evidence.
Consider the timing hypothesis. What if the twelve-address cluster routes through HumidiFi primarily during periods when market volatility is elevated, when Solana's public pools are flooded with noise and transaction failures spike? In that scenario, HumidiFi's apparent superiority could be a function of trading regime, not venue quality. The cluster looks better because it is trading in better conditions, not because HumidiFi is a better venue. The study's data structure allows for this confound and does not fully resolve it.
Consider the behavioral hypothesis. The twelve addresses may run a fundamentally different strategy: longer holding windows, more selective entry timing, different position sizing. If their strategy is generally superior, and HumidiFi happens to be the venue they prefer when executing that strategy, the venue association is a shadow of a deeper strategic difference. The study compares within-cluster performance against HumidiFi versus outside HumidiFi, which controls for strategy to some degree. But the contrast is not a controlled experiment. It is a natural experiment with clean data and messy confounds.
None of this invalidates the finding. It constrains it. The correct interpretation is: among the MEV-like addresses sampled, those touching HumidiFi achieved dramatically better risk-adjusted outcomes in the October-to-November 2025 window than the same addresses achieved elsewhere. Whether that gap reflects venue quality, congestion timing, strategic selection, or a feedback loop is undetermined. The white-box verification needed to answer that question rests on HumidiFi's owners publishing their code. They have not. Until they do, the study remains a dark-pool canary, not a trading manual.
This is also why the paper's methodological caution is strategically important. A less disciplined analysis would have produced a clickbait conclusion. The authors controlled themselves. In doing so, they preserved the credibility of the finding for the audience that matters: professional infrastructure teams that can independently probe HumidiFi and decide whether to include it in their routing graphs.
The Tokenless Cash Machine
There is no HumidiFi token. No governance proposal. No emissions schedule. No community treasury. From a token-economics perspective, the protocol is a zero. And that, perversely, is the most economically interesting part.
Most DeFi protocols issue tokens to bootstrap liquidity. The token is a subsidy, a way to pay early LPs for bearing risk before the network effect materializes. HumidiFi flips that model. Because it is proprietary and closed, it does not need to subsidize liquidity acquisition. The liquidity is either self-supplied by the operator or sourced from a small number of counterparties who receive a private benefit: access to profitable routing. The absence of a token is not a missing feature. It is an anti-feature, a deliberate elimination of the transparency mechanisms that public token models incidentally create.
The WSOL dimension of the study matters here. All outcomes are measured in Wrapped Sol. In a bull phase, WSOL-denominated profits amplify in fiat terms. In a bear phase, they vanish entirely even when the strategy succeeds. The cluster's 62.3% positive-outcome rate says nothing about US-denominated returns. It says only that, in SOL terms, the cluster's WSOL balance usually grew after a HumidiFi interaction. That is an important distinction for any allocator evaluating MEV strategies as a portfolio component.
There is also a transfer-analysis question that the paper does not address. If HumidiFi's edge is real, where does the profit come from? In an efficient market, every winner implies a loser. Candidates include public-venue LPs who suffer adverse selection when better-informed trades route elsewhere, retail traders who receive unfavorable fills on public venues because intelligent flow has left those pools, and arbitrageurs who lose races they do not know they are running. The paper's data structure cannot assign losses to specific groups. But the distributional logic is inescapable. An information edge of 3x does not materialize out of a vacuum.
The Stratification of Solana's Execution Layer
Solana's DEX ecosystem is not a monolith. The study makes that explicit. There is a public aggregation layer dominated by Jupiter. There are public liquidity venues like Raydium and Orca. There is a proprietary layer, HumidiFi and, presumably, venues like it that no one has yet indexed. And there are bot services like Trojan and SolanaMevBot that package execution for end users with varying degrees of technical sophistication.
The twelve-address cluster sits at the apex of this stratification. Its members route through Jupiter heavily, the study notes their traffic is Jupiter-dependent, but they know that Jupiter's route candidates omit HumidiFi. They are, in effect, using the public aggregator for everything except the one destination that matters most. Jupiter's visibility is scaffolding; their private knowledge is the load-bearing wall.
The comparison with traditional finance is impossible to avoid. Dark pools in equity markets offer institutional traders reduced market impact at the cost of transparency. Regulators demand that dark-pool operators provide certain disclosures, which limits the opacity advantage. On-chain, no such disclosure regime exists. HumidiFi-style venues can operate as fully anonymous liquidity facilities, invisible to every participant except the few who are permitted or smart enough to find them.
WSOL's dominance in the sample is another microstructural tell. The study found Wrapped Sol appearing in 70.90% to 99.94% of transactions across MEV-like clusters. In Ethereum's MEV economy, WETH plays the same role: the numeraire for arbitrage, the marginal asset for liquidity provision, the settlement unit for value extraction. WSOL's equivalent status makes Solana's MEV economy structurally similar to Ethereum's, even though Solana's parallel execution architecture was supposed to reduce MEV risk. The protocol design changed the mechanism; it did not change the incentive.
The Pump.fun cluster, with its 80.9% concentration, completes the picture. Solana's MEV population has split into at least two distinct sub-ecologies: traditional DEX arbitrage centered on WSOL and the major venues, and meme-coin launch racing centered on Pump.fun and its periphery. They use different strategies, different tools, and different risk profiles. They do not compete directly. The twelve-address HumidiFi cluster is a specialist in the first ecology. That specialism is precisely why it found an edge the generalist market missed.
For Jupiter, the strategic implication is uncomfortable. As the dominant aggregator, Jupiter's routing algorithm determines what "best execution" means for millions of retail users. If Jupiter does not include HumidiFi in its route candidates, then Jupiter users are systematically receiving inferior prices whenever HumidiFi holds the true best quoted depth. The study does not quantify this gap in price-impact terms. But the WSOL outcome differential is a proxy. The public aggregator's route graph has a blind spot, and the blind spot is being exploited by a small, sophisticated population.
Propagation: From Liquidity-Driven to Routing-Driven
This study's deepest industry consequence is directional. It accelerates Solana's DeFi evolution from a liquidity-driven market, where total value locked and volume were the primary success metrics, toward a routing-driven market, where the selection of execution paths determines realized returns.
For infrastructure providers, that is an unambiguous positive signal. Data-indexing services like Solscan, Dune, and The Graph will face growing demand for finer-grained program identification and behavioral labeling. Address-tagging services become more valuable when the difference between a 21% and a 62% positive-outcome rate depends on identifying the right program ID. Transaction simulators gain importance as tools that allow new entrants to test HumidiFi-style routes without committing capital. The entire middleware stack benefits from the discovery that route selection is a performance variable, not an afterthought.
For liquidity providers on public venues, the signal is negative and direct. If professional flow increasingly migrates to closed pools, public-venue LPs will find themselves structurally disadvantaged. They are already fighting adverse selection; now they are fighting an information gap they cannot observe. This can create a feedback loop: as intelligent flow abandons public pools, public pool spreads widen, which pushes even more flow into proprietary venues, which further degrades the public venues' fills. In an unforgiving reading, this is the start of a classic adverse-selection death spiral for public LP capital.
For institutional trading teams, the study reads as an invitation. The discovery that a twelve-address cluster can sustain a near-3x edge through a closed venue invites replication. Teams with engineering resources will deploy their own proprietary pools, whitelist their own traders, and internalize their own flow. The likely consequence is a multiplication of HumidiFi-style venues across Solana, each privately guarded and each eroding the market-share of public liquidity. This is not a prediction of fraud. It is a prediction of incentive-driven market microstructure evolution.
The risk in that trajectory is that on-chain value extraction does not disappear. It migrates from transparent front-running to opaque private extraction. Retail users cannot defend against a route they cannot see. The chain will continue to publish every transaction, so the extraction will remain theoretically discoverable, but the burden of discovery now falls on unsophisticated participants rather than on the infrastructure that is supposed to protect them through transparent routing.
The Regulatory Shadow
If an equity-market structure analyst at the SEC read this paper, they would recognize the anatomy instantly. A private venue, inaccessible to the public, providing preferential execution to a connected group; that is a best-execution violation waiting for a plaintiff. On-chain, the legal framework is murkier. But the interpretive frame is converging.
RFIA classified digital assets in 2022. MiCA imposed licensing requirements on exchanges and custodians. Neither regime directly addresses MEV extraction or proprietary AMM access. That does not mean the regulatory risk is zero. It means the risk is unmodeled. The paper's findings provide exactly the kind of microstructural evidence that regulators use to build novel enforcement theories. If a subsequent study quantifies retail losses from HumidiFi-style routing disadvantages, the argument for consumer-protection intervention in DEX base layers becomes more concrete.
I am not predicting immediate enforcement. I am identifying the trajectory. The regulatory conversation around crypto has shifted from "is this a security" to "does this market operate fairly." The latter question is far harder to litigate and far easier to generate evidence for. This paper is evidence. No legal team has built a case on it yet. But the data is public, reproducible, and timestamped. It is waiting for the lawyer who knows how to use it.
The dark-pool analogy also carries an obligation. For professional trading teams considering replicating the HumidiFi cluster's strategy, the reputational cost of building a private AMM that extracts from retail flow is nontrivial. Traditional institutions have spent years being fined for dark-pool misconduct. On-chain, the equivalent abuses are harder to police and easier to expose. The chain does not forget. Solscan annotations persist. The replication package is forever.
The Risk Matrix
Let me be systematic about what can break.
The most immediate risk is edge decay. The paper is now public. The replication package is public. The program ID of HumidiFi is public. Any data-savvy team can rebuild the cluster's trading rules, query HumidiFi's pools, and begin competing for the same deviations. If HumidiFi's contract is open to all callers, this competition will compress the 62.3% positive-outcome rate toward the environment's average within a quarter. If the contract restricts access, through whitelists, signatures, or data-provider gates, the edge persists for the privileged. The study cannot tell us which case applies. Its own hedging language suggests uncertainty.
The second risk is method blindness. The paper's metric is unrefined. WSOL balance changes include fee payments, priority fees, and capital additions. The authors explicitly warn that the measure is not a fully cleaned strategy return. An LP position that increases in value during a SOL bull phase records a positive WSOL balance change even if the strategy itself was poorly executed. Distributional windows matter. The 62.3% statistic could shift significantly if fees and timing were fully isolated.
The third risk is liquidity fragility. A closed venue's depth is opaque. If HumidiFi's pool is shallow, the cluster's success is limited to small order sizes. A replicator deploying larger capital would encounter slippage that the study's trade-size distribution cannot illuminate. The strategy may not scale. Alternatively, if the pool is deep but the LP base is the operator's own capital, the operator's risk tolerance governs the venue's survival. A single large drawdown in the operator's proprietary book could close the venue, taking the cluster's edge with it.
The fourth risk is narrative contagion. This study, amplified, can feed Solana's existing anxiety about robot dominance. CryptoSlate has previously reported on sandwich attacks, bot congestion, and subsidized professional order flow. The study's neutral framing is a choice. A less disciplined writer can convert the finding into "Solana retail is being systematically drained by dark-pool bots." That narrative has real sentiment value. If it infects SOL's social layer, the market impact may reflect the narrative rather than the mechanism.
The fifth risk is strategic misreading. A trader who treats this paper as a route to easy profits will likely lose money. The cluster's edge depends on execution conditions, latency, calldata structure, program-interaction sequence, gas strategy, conditions the paper does not fully document. Replicating an outcome without replicating the mechanism is the oldest failure mode in trading. I watch this industry repeat that mistake every cycle.
A Contrarian Interlude
Let me steelman the other side before issuing a verdict.
First, the transparency paradox works in Solana's favor. On a closed centralized database, this edge would never be noticed. On Solana, the transactions are publicly indexed, the program ID is identifiable, and the academic paper is reproducible. The outcome asymmetry, however uncomfortable, was discoverable by anyone with the patience to look. That discoverability is itself a corrective mechanism. Markets that can observe their own asymmetries are more likely to correct them than markets that cannot.
Second, proprietary execution venues are not inherently malicious. In traditional finance, internalization and crossing networks serve a legitimate function: reducing market impact for large orders that would otherwise move public prices. If HumidiFi's advantage derives from better execution quality, not from exploitation, the protocol might actually improve aggregate market efficiency. Small orders routed to public venues would be unaffected; large orders that would otherwise wreak havoc on a public book find a stable counterparty. The study cannot distinguish efficient crossing from extractive front-running. We are not entitled to assume the worse case.
Third, the competitive dynamic is healthy. The discovery of HumidiFi-style edges incentivizes infrastructure teams to build better routing, to scan for unlisted venues, and to treat execution quality as a first-class product dimension. Jupiter's response to this paper will be telling. If it expands its route graph to cover closed and semi-open venues, the retail execution gap narrows. If it does not, the paper will become a recurring critique of the aggregator model. Competitive pressure, not regulation, is the shortest path to fairer execution.
Finally, the Wolf of Wall Street framing obscures a mundane truth: the twelve-address cluster is doing what every sophisticated trader in every market in history has done. It found an information edge and exploited it. The real failure is structural, not moral. A network that promised permissionless equality of access delivered permissionless inequality of knowledge. The edge was neither stolen nor hacked. It was found. The response, if there is to be one, should focus on closing the information gap, not punishing the information holder.
The Verdict
The proof is complete; the doubt is obsolete. Not because we know exactly how HumidiFi generates its edge, but because we have established a new baseline. Execution on Solana is not equal. It cannot be assumed as equal. Routing decisions are now as important as strategy decisions, and both are more important than governance theater.
The lesson for builders is unambiguous: open the code. Publish the audit. Broadcast the liquidity. Any protocol that chooses opacity in an open network is choosing to be a structurally extractive layer, no matter its stated intentions. The code whispers secrets either way. In this case, the secret is a 3x edge in a shadow pool.
I do not trust; I verify the hash. HumidiFi's hash is unavailable. So my verdict is: unverified, structurally suspicious, and very likely profitable. If that ambiguity bothers you, it should. The chain never lies; it simply hides in plain sight.