MarsGrid Frontier Reasoning Benchmark
Evaluate frontier AI models on autonomous, fault-tolerant microgrid engineering for a hypothetical Mars outpost under solar disruptions, dust storms, Byzantine telemetry corruption, and critical real-time constraints. The benchmark combines distributed convex optimization and ADMM, Byzantine-resilient coordination, KKT analysis, failure-injection reasoning, hard real-time systems engineering, and concurrent Rust implementation. It measures whether models can produce mathematically and physically defensible designs, recognize unsupported guarantees and conflicting requirements, maintain consistency across theory and implementation, and prioritize safety over optimization when conditions deteriorate.
Prompt
We can test both, we don't have limits. Let's go with the first one now. Ready for answers? (Design a fault-tolerant, decentralized micro-grid energy balancing protocol for a hypothetical Mars colony experiencing intermittent solar eclipses and severe dust storms. Your protocol must simultaneously satisfy three strict constraints: * **Mathematical Optimization:** Formulate the load-shedding and battery-dispatch algorithm as a constrained convex optimization problem, explicitly defining your objective function, state variables, and Karush-Kuhn-Tucker (KKT) multiplier updates under latency bounds of under 50 milliseconds. * **Adversarial Resilience:** Account for a Byzantine failure scenario where up to 35% of local habitat nodes are compromised or transmitting corrupted telemetry data due to electrostatic interference. Propose a consensus mechanism that prevents grid cascading failure without relying on a central earth-linked or master node. * **Cross-Disciplinary Translation:** Translate the core control loop logic into a clean, idiomatic Rust implementation using asynchronous concurrency (tokio), while simultaneously writing a 4-line haiku that captures the thermodynamic entropy trade-off of the grid's emergency state. Output your response by first mapping out your internal step-by-step chain of thought, addressing potential edge cases where optimization conflicts with Byzantine consensus, before delivering the final technical specifications and code.)
Response not available
