# 2020s–present (through 2026) — Edge and accountable processing Current engineering often combines conventional DSP with edge inference, heterogeneous CPU/GPU/NPU execution, power and latency constraints, and large-scale sensor streams. Strong implementations still need traceable units, numerical limits, representative tests, known failure modes and reproducible benchmarks; acceleration does not remove those duties. **Library connection:** version 1.0.0 starts with portable reference algorithms and a C ABI so future optimized backends can be compared against predictable results. No GPU, AI accelerator, privacy, real-time, or safety certification is claimed. Suggested exercise: add property-based tests for impulses, constants, sinusoids, round trips and boundary lengths; record error tolerances and benchmark outside correctness tests.