Dr. Johnnatan Messias

This paper investigates how crypto airdrops have transformed from marketing campaigns into a new form of platform-mediated finance. By analyzing over 37,000 YouTube videos and Telegram communities across seven major blockchain projects, we show that token value is increasingly created through digital labor, community participation, and platform dynamics rather than speculation alone. We introduce the concept of post-speculative finance, highlighting how blockchain ecosystems rely on social media, governance, and user-generated content to build legitimacy and sustain digital economies.

Blockchain transparency and immutability come at the cost of an ever-growing ledger, which threatens long-term decentralization. We introduce state-invariant transactions—transactions that do not affect the blockchain state beyond paying transaction fees—and show that they represent a form of on-chain spam that wastes execution, storage, bandwidth, and blockspace. Analyzing nearly 1.4 billion such transactions across Ethereum, Optimism, and Base, we find they are particularly prevalent on Layer-2 networks, where speculative MEV is the dominant source, alongside significant malicious activity such as address poisoning. Our results suggest that reducing state-invariant transactions could lower blockchain resource consumption, decrease transaction costs, and mitigate blockchain abuse.

We study the designs and implementations of 48 public, actively used DAOs deployed on Ethereum and identify how three key governance mechanisms — token registration, staking, and delegation — systematically reinforce centralization of voting power, revealing inherent trade-offs between decentralization, security, and usability.

We analyze 48 public Ethereum-based DAOs to uncover trust and transparency trade-offs in governance design, discovering a new class of governance attacks that exploit fundamental design choices in DAO governance mechanisms — even in bug-free implementations.

We identify 500,000+ previously untapped arbitrage opportunities across Arbitrum, Base, Optimism, and ZKsync, demonstrating that non-atomic arbitrage on rollups is economically meaningful and structurally distinct from L1 Ethereum — calling for new MEV extraction strategies.

A formal modeling approach using Stochastic Petri Nets to assess operational behavior of ZK-Rollups in cloud environments, calibrated using empirical data from zkSync Era. We show that increasing L2 transaction volume can reduce daily operational costs by up to 25%, with a significant cost-efficiency vs. responsiveness trade-off.

Using Stochastic Petri Nets, we show that increased adoption of Layer-2 transactions can increase system throughput by up to 20%, while latency may increase by more than 100% with larger batches — revealing a fundamental performance trade-off in ZK-Rollup design.

The first large-scale empirical study of Timeboost, analyzing 11.5M+ express lane transactions and 151K auctions. We find express lane control is highly centralized (two entities win 90%+ of auctions), ~22% of transactions are reverted, and secondary markets have collapsed — showing Timeboost fails to deliver on fairness, decentralization, and spam reduction.

An empirical study combining on-chain data from five major protocols with off-chain DAO forum discussions. Using LLMs to extract governance interests, we show delegations are frequently misaligned with token holders' expressed priorities and that ranking-based interfaces exacerbate power concentration.

The first comprehensive empirical study of nine major airdrops. Up to 66% of tokens are rapidly sold, often in recipients' first post-claim transaction, driven by "airdrop farmers." We identify common design pitfalls and propose actionable guidelines for more effective airdrop strategies.

An in-depth empirical study of Compound and Uniswap governance protocols, analyzing 370+ governance proposals from inception to August 2024. As few as 3–5 voters were sufficient to sway the majority of proposals. The cost of voting disproportionately burdens smaller token holders, and strategic voting behaviors further distort governance outcomes.

A Stochastic Petri Net model to evaluate the feasibility of ZK-Rollups by analyzing impact on throughput and latency. Increased adoption of Layer-2 transactions can enhance throughput by up to 20%, while latency may rise by more than 100% with larger batches — revealing a fundamental performance trade-off.

On certain days inscriptions accounted for nearly 90% of transactions on Arbitrum and ZKsync Era, and 53% on Ethereum, with 99% involving meme coin minting. ZKsync and Arbitrum saw lower median gas fees during these surges — ZKsync Era showed a greater fee reduction than optimistic rollups.

We curated a dataset from ~1 year of activity extracted from a ZKsync Era archive node, made freely available to the research community. The paper details its creation, showcases example analyses, and discusses future research directions for Layer-2 data-driven research.

A theoretical and empirical study of LSTs on AMMs. While trading fees often compensate for impermanent loss, fully staking is more profitable in many pools — raising questions about the sustainability of current LST liquidity allocation to AMMs.

We introduce a theoretical framework for Maximal Arbitrage Value (MAV) and conduct an empirical study of arbitrage between SyncSwap (zkSync Era) and Binance. Total MAV from July–September 2023 in the USDC-ETH pool: $104.96K (0.24% of trading volume).

We characterize the lack of contention and prioritization transparency in Bitcoin and Ethereum from private relay networks, showing private transactions are prevalent, facilitating miner collusion and overcharging users — with critical implications for blockchain stability.

We audit the Bitcoin blockchain and present statistically significant evidence of mining pools deviating from fee-per-byte norms to accelerate selfish transactions or those with dark-fee payments — calling for urgent discussion on neutrality norms in blockchain transaction ordering.

We study efficiency in proof-of-work blockchains with non-zero latencies, proposing a model for both P2P and coordinated mining pool settings. We derive closed-form expressions for efficiency and show that under a natural consistency condition, overall system efficiency in P2P is higher than in coordinated settings.

In a month-long investigation of Bitcoin we reveal that congestion is typical and commit times exhibit significant variance. Although fee-per-byte is the widely accepted "norm," we show that miners significantly delay a fraction of transactions — deviations that undermine the utility of blockchains for fair ordering.

A large-scale analysis of misinformation dissemination in WhatsApp public political groups during the 2018 Brazilian elections — identifying key sources of fake images, propagation patterns across groups, and network structures that enable misinformation spread.

We examine Russian IRA ad campaigns prior to the 2016 U.S. elections, showing how Facebook's advertising infrastructure enabled efficient targeting of divisive political content at vulnerable sub-populations — and calling for greater transparency in political ad targeting.

We make the first attempt to quantify and explore demographic biases in crowdsourced recommendations — finding that very different topics are popular among different demographic groups and that there is a systematic bias toward particular demographics in trending topic selection.

A web-based system that helps researchers and journalists explore content shared on WhatsApp public groups in Brazil and India. The tool monitors images, videos, audio, and text, displaying most-shared content per day. It was used to monitor content during the 2018 Brazilian general election and was a major source for estimating misinformation spread.

We propose a generalizable search bias quantification framework that measures political bias in ranked search results and decouples bias from input data versus the ranking system. Applied to 2016 U.S. Presidential primaries on Twitter, we find both input data and ranking system shape final search output bias.

Using Face++ image processing algorithms to identify gender and race of U.S. Twitter users, we show that users identified as White and male tend to attain higher positions in terms of follower count — demonstrating that social media perpetuates offline demographic inequalities.

The first in-depth characterization of news spreaders in social media. Males and white users tend to be more active in sharing news, biasing the news audience toward the interests of these demographic groups — with implications for personalized news digests.

We characterize language usage across demographic groups in Twitter, extracting linguistic features from 6 categories. Clear differences emerge in writing styles and topics of interest across both gender and race lines among U.S. Twitter users.

The first attempt to quantify demographic biases in Twitter's crowdsourced trending topics. A large fraction of trends are promoted by crowds whose demographics differ significantly from the overall Twitter population, with certain groups systematically under-represented.

A large-scale measurement study showing that more than 28% of 6-year old public tweets are no longer accessible today. Even when users delete data, current mechanisms leave traces of residual activity — and we show that significant information can be recovered from this residual data.

The first study comparing 14 sentence-level sentiment analysis methods adapted for Android OS, measuring memory, CPU, and battery consumption. We identify methods that run efficiently on mobile devices as well as those that cannot be deployed due to excessive memory use.

A large-scale measurement study of how Twitter users control longitudinal exposure of their public tweets. Users face two fundamental problems: residual activity traces persist even after deletion, and withdrawal mechanisms attract undesirable attention to deleted content.

Using "places lived" data from millions of Google+ users, we study migration clusters — groups of countries individuals have lived in — going beyond bilateral flow models. Migration clusters of country triads cannot be identified using bilateral flow data alone.

A comparative study of 17 sentence-level sentiment analysis methods adapted for Android OS, measuring performance in terms of memory usage, CPU usage, and battery consumption — providing a practical guide for developers building mobile NLP applications.

More than 28% of 6-year-old public tweets on Twitter are no longer accessible today. Even after deletion, residual activity traces remain — allowing significant recovery of deleted content. We propose an exposure control mechanism that eliminates information leakage via residual activities.

Using Twitter bot accounts that interact with real users, we show it is possible to become influential in systems like Klout and Twitalyzer through very simple strategies — suggesting these influence measurement systems should be revised to avoid rewarding automatic activity.