# Numerical safety and known boundaries ## Input validation The C routines check required pointers, positive lengths, transform modes, capacities, and relevant size arithmetic. Numeric routines reject NaN and infinite input samples where documented and check for non-finite results. Callers must still supply valid allocated buffers of the claimed lengths; C cannot verify the actual size or lifetime of arbitrary pointers. ## Precision The public sample type is double. Some accumulations use long double, whose range differs by platform. Floating-point results can vary slightly among architectures, compiler flags, math libraries, and evaluation order. Tests should compare floating-point results with reasoned tolerances rather than require exact bitwise identity for transcendental operations. ## Aliasing The moving average and convolution contracts require non-overlapping input and output buffers. The DFT also requires separate, non-overlapping arrays. The FFT is explicitly in place. The routines do not attempt general alias detection. ## Operational boundaries Version 1.0.0 does not promise hard-real-time behavior, bounded execution time, thread-safe shared state guarantees for future stateful processors, fixed-point arithmetic, SIMD/GPU acceleration, denormal handling policy, or safety certification. The current routines are stateless and use no shared mutable library globals, but callers are responsible for synchronizing their own buffers and processing pipelines. Before using results in regulated, medical, flight, automotive, or other safety-critical equipment, the system owner must create a domain-specific validation plan and independently verify the full acquisition and processing chain. These baseline utilities alone are not certified for such use.