## TODO

### C Rewrite Polish
* Win32 threads branch in `src/parallel.c` (Windows currently falls back to serial)
* Persistent thread pool instead of per-gate fork-join in `psi_parallel_for` (avoid thread spawn overhead)
* Verify the x86 SIMD paths (`apply_avx2`/`apply_avx512` in `src/maths/simd.c`) on real AVX2/AVX-512 hardware — only the NEON path has been run
* `tools/check-style.sh` (grep-based) for the rules clang-format/clang-tidy can't enforce: no `typedef` on structs, `double`-only (no `float`), stdint over `int`/`long`, `sizeof *obj` allocation idiom
* `DensityMatrix` Display/Debug pretty-printer (skipped during the port; cosmetic)

### Noise Channels & Open Quantum System Modelling
* Depolarising channel
* Amplitude damping
* Phase damping
* Thermal relaxation (T<sub>1</sub>/$T_2$)
* Readout noise
* Kraus operator framework
* Pauli / Lindblad evolution engine
* Stabiliser noise models

### Density Matrix Backend
* Optimised density matrix propagation
* Hybrid wave-function / density-matrix simulation modes

### GPU Offload, Caching & Scheduling
* State caching on GPU
* Tensored simulation on device
* Batched execution engine
* Device-side gate fusion

### Tensor Network Simulation Backend
* Tensor slicing
* Contraction ordering
* Edge-cut optimisation

### Analogue Evolution
* Suzuki–Trotter time-evolution models