Ipak Systems
Bringing order to complex systems.
Ipak Systems is an applied complexity science research practice. We bring methods from physics, statistical mechanics, and evolutionary modeling to decentralized financial systems - calibrating parameters, modeling user behavior empirically, and identifying the regime boundaries that separate normal operation from failure. Our current focus spans lending protocols, DEXs, perpetuals, and the markets that connect them.
What complexity science reveals in DeFi systems
Behavior-aware modeling
User behavior is an empirical input in our models, measured from an on-chain history rather than assumed. Borrowers, lenders, liquidators, and LPs are rarely rational and each have characteristic response times, sensitivity to costs, and decision rules. Models calibrated to the actual behavior of a protocol's users, not a representative agent, capture dynamics that matter at the parameter level.
Tail-event simulation
Stress tests run against scenarios drawn from outside the historical record as well as inside it. Heavy-tailed processes, extreme value theory, and scenario synthesis let us evaluate parameter sets against the regime shifts they may eventually have to survive.
Agent-based and evolutionary modeling
Liquidation cascades, MEV interaction, and curator-vs-vault dynamics involve many agents whose strategies adapt over time. Multi-agent simulation with evolutionary strategy update captures these dynamics where representative-agent models cannot.
Network and coupling analysis
Shocks propagate across protocols through shared collateral, shared oracle paths, and overlapping liquidator pools. We treat the multi-protocol environment as a network and analyze how configuration at one node changes the failure surface of its neighbors.
Phase transitions and stress regimes
The line between "normal" and "stressed" is mechanical, not visual. We identify the parameter moves that put a protocol one step away from regime change, and the monitoring rules that detect approach to that boundary before it is crossed.
How this shows up in our work
Parameter calibration as a joint system
Every protocol has a solvency condition - an inequality between what it can absorb and what it might be asked to absorb under stress. The parameters governing that protocol are jointly bound by this condition: caps and liquidation parameters in lending, margin and insurance-fund parameters in perps, curve and fee parameters in AMMs. We derive parameter sets from the binding inequality across all routes simultaneously, integrating asset analysis, liquidity structure, and behavioral calibration. Changing any one parameter forces re-derivation of the others.
Simulation environments built per system
We test parameter sets and model changes against scenarios drawn from outside the historical record as well as inside it. The toolkit spans heavy-tailed price and depth processes calibrated to observed distributions and extrapolated using extreme value theory; multi-agent simulation with evolutionary strategy update for adaptive dynamics like liquidation cascades and MEV competition; network simulation across coupled protocols where shocks propagate through shared collateral and oracle paths; and phase-transition analysis to identify the regime boundary a given configuration sits closest to. We build custom simulation environments where standard frameworks don't fit the system.
Behavioral and network research
Some questions don't fit cleanly into a calibration cycle: how a protocol's users actually segment by behavior, how shocks propagate across coupled markets, what the observable signatures of pre-failure look like, and where on the configuration surface a protocol sits relative to its nearest regime boundary. We take these on as standalone engagements, drawing on tools from network science, agent-based modeling, and statistical mechanics.
Full-cycle modeling
Every engagement runs from qualitative investigation through mathematical formalization to deployment-ready parameter sets or research output. Methodology is documented openly and the working research is published as it develops.
See Services for current engagement structure, or read the research.
Get in touch — research@ipak.systems