Defines an LSTM-based model with attention mechanism for classifying assembly instruction sequences into operator types (e.g., conv, add, relu). The component is self-contained, depending only on PyTorch and standard libraries, and demonstrates reusable sequence classification architecture applicable to other binary analysis tasks.

Implements a symbolic execution engine for recovering tensor shapes from execution traces. It handles memory and register states, supports linear arithmetic constraints, and is decoupled from the specific decompilation workflow. This engine demonstrates non-obvious skill in constraint solving and state management, making it reusable for other binary analysis tasks requiring symbolic reasoning.

Provides dynamic taint analysis for filtering instruction traces to identify instructions affecting specific memory regions. It implements forward and backward taint propagation with custom handling for x86-64 instructions (mov, arithmetic, logical, xmm-specific). The component demonstrates reusable taint analysis patterns with global state management and can be adapted for other dynamic binary instrumentation scenarios.