kinexys by j.p. morgan
i like measurement problems
maximal extractable value (mev) refers to the value that can be extracted from block production beyond standard block rewards and transaction fees by including, excluding, and changing the order of transactions. in theory, mev accrues entirely to validators since they control transaction inclusion, exclusion, and ordering; in practice, most mev is extracted by independent network participants referred to as "searchers" who detect profitable mev opportunities and submit these transactions to validators, typically paying high gas fees for inclusion. assuming searchers are economically rational, the gas fee that a searcher is willing to pay will be an amount up to 100% of the extracted value.
mev takes many forms; flashbots even lists a job board detailing mev opportunities (and new constantly emerge as new protocols develop and old protocols upgrade). this environment presents an interesting measurement problem: what transactions should be considered mev and how should we classify them? there is a lot of literature on this taxonomy problem and the team at sorella labs has done an amazing job at actually building measurement systems for these types of transactions on ethereum, paying particular attention to decentralized exchange (dex) arbitrage, sandwich attacks, just-in-time (jit) liquidity, and lending protocol liquidations.
however, this measurement problem becomes increasingly difficult on low-latency, high-throughput blockchains like solana or aptos. with subsecond blocktimes and low transaction fees, spam becomes a significant problem and a significant share of that spam consists of failed mev transactions. i am building pono to detect successful and failed mev transactions. this project is motivated by three initiatives: (a) to bring to light how much mev is extracted on the block level and how much is accrued to the searchers submitting the transactions and the validators including them; (b) to understand how many transactions are attributable to failed mev opportunities and how much blockspace they consume; and (c) to study mev around network and protocol upgrades as well as market events to assess their impact if any.
kalshi and polymarket have now processed billions of dollars in trades, creating liquid markets for contracts tied to elections, macroeconomic releases, sporting events, and other real-world outcomes. while these markets are compared to traditional financial derivatives, binary contracts differ fundamentally from traditional options: they converge to a payoff of either 0 or 1 at expiration, changing both the geometry of the implied volatility surface and the arbitrage relationships between related contracts.
binary options and implied volatility have been studied extensively in quantitative finance, but prediction markets introduce a new measurement challenge: whether the pricing frameworks developed for traditional derivatives remain valid in a nascent asset class with binary payoffs, sparse liquidity, and event-based underlyings. when prediction markets offer contracts related to traded financial assets, do their prices agree with prices and implied probabilities from traditional markets? which market incorporates information more efficiently? do overlapping prediction markets satisfy static no-arbitrage conditions or do market microstructure effects produce persistent pricing inconsistencies?
i am building a dataset that aligns historical prices from kalshi and polymarket with related instruments from traditional financial markets, including equities, interest rates, and derivatives, so that i can study cross-market price discovery, calibration, and arbitrage. i also ask whether prediction markets require new pricing frameworks.
chaos-based cryptography has long been motivated by the intuition that properties of chaotic dynamical systems (sensitivity to initial conditions, ergodicity, and high kolmogorov-sinai (ks) entropy) should translate into cryptographic security. but existing chaos-based constructions have repeatedly failed under cryptanalysis because dynamical complexity does not imply computational hardness. i ask whether chaotic systems can be rigorously developed into cryptographic primitives by identifying conditions under which dynamical unpredictability and computational hardness can coexist.
the core challenge is bridging two notions of difficulty: the informal hardness of chaotic dynamics and the computational hardness required for secure cryptographic constructions. i study whether algebraic dynamical systems can retain desirable chaotic properties while being grounded in established hardness assumptions, particularly lattice problems like ring learning with errors (rlwe) and the module short integer solution (msis).
one direction examines nonlinear maps over lattice-based algebraic structures, such as $\varphi(a)=a^2+e$ over the ring $R_q=\mathbb{Z}_q[x]/(x^n+1)$, where $e$ is sampled from an rlwe error distribution. the central questions are whether such maps can exhibit useful dynamical properties, including ergodicity and high ks entropy, while remaining in parameter regimes where the underlying lattice problems are hard; and whether recovering or predicting elements of the resulting orbit $\mathcal{O}(a_0)={a_0,\varphi(a_0),\varphi^2(a_0),\ldots}$ can be reduced to solving an established computational problem.
a related direction studies the relationship between ergodicity and computational hardness in other algebraic dynamical systems. anashin's characterization of ergodic 1-lipschitz maps over $\mathbb{Z}_p^n$ provides conditions for uniform mixing, but does not address whether these systems are difficult to invert. i investigate whether polynomial maps over $p$-adic domains can simultaneously achieve strong dynamical properties and computational hardness, and whether hyperbolic toral automorphisms of the form $T_M(x)=Mx \bmod q$ can connect rapid mixing dynamics with hardness assumptions such as msis.
another perspective comes from cryptographic hash functions, which achieve apparent randomness and unpredictability through computational hardness rather than explicit dynamical structure. when iterated over finite state spaces, hash functions induce deterministic dynamical systems with eventual cycles, yet their behavior is often modeled as random mappings. this contrast highlights a fundamental distinction between dynamical randomness and computational randomness: chaotic systems provide measurable properties such as entropy and mixing rates, while cryptographic primitives provide security through hardness assumptions. a central question is whether these two notions of unpredictability can be combined in a single construction.
at a more practical level, i explore nonlinear variants of lattice cryptography, where the linear transformations underlying lwe are replaced by nonlinear chaotic maps. even without an immediate security reduction, these constructions provide a framework for studying whether introducing nonlinear structure can improve resistance against attacks that exploit the algebraic regularity of existing schemes. underlying all of these approaches is a fundamental discretization problem: continuous chaotic systems often collapse into short cycles when projected onto finite rings, making their apparent unpredictability cryptographically useless. identifying discrete dynamical systems with long trajectories, efficient computation, and provable hardness remains the central challenge.