Deterministic Frame Stability via DAG Scheduling

This paper, titled "Deterministic Frame Stability via DAG Scheduling," provides a formal architectural account of how the C.A.T. engine treats stable frame delivery as a provable design invariant rather than a byproduct of hardware performance. The research argues that frame stability is achieved through a formal control stack consisting of a timing system that advances a monotone frame tick, a weighted render DAG, a greedy job scheduler, and a bounded swapchain queue. The core contributions of the paper include: A Discrete Timing Contract: The engine advances time on a discrete lattice, clamping delta times to ensure a predictable "target-rate frame contract". Formal Render Graph Compilation: The system transforms a declarative list of passes into a topological execution order, performing resource culling and lifetime-based memory aliasing to reduce churn and spikes. Graph-Theoretic Job Scheduling: Parallel work is managed through a greedy list-scheduler that uses dependency-satisfied ready sets and work-stealing to bound execution time by the weighted critical path of the job graph. Swapchain Back-Pressure: The paper formalizes the Vulkan swapchain as a bounded producer-consumer buffer, proving that this boundary prevents the CPU from outrunning the display cadence. Physics Stage Regularization: Simulation maintenance is modeled as a periodically sampled DAG, allowing for amortized costs and parallel safety through read-only ECS snapshots. The paper synthesizes these subsystems into an end-to-end frame bound equation, demonstrating that frame duration is limited by the longest declared critical paths and explicit synchronization points. By narrowing the runtime surface compared to general-purpose engines, the authors show how jitter can be transformed from a structural default into a diagnosable, explainable outlier. The result is an engine architecture where the frame is not just fast, but scheduled, bounded, and back-pressured by construction.