# IO reports

Dated records retain their original workload, backend, source revision and
acceptance limits. They are evidence, not a statement of current release speed.

- [QEM conversion and complete lossless verification, 2026-10-23](2026-10-23-qem-verification.md)
- [QEM index size and CUDA loading, 2026-10-23](2026-10-23-qem-layout-cuda-profile.md)
- [QEM compression compared with HDF5, 2026-10-23](2026-10-23-qem-compression.md)
- [Bounded MPS HDF5 loading, 2026-10-23](2026-10-23-mps-hdf5-pipeline.md)
- [Acquisition-worker loader preparation, 2026-10-23](2026-10-23-loader-startup.md)
- [Native Swift/Metal IO benchmark, 2026-10-23](2026-10-23-io-swift-metal.md)
- [Python MPS IO benchmark, 2026-10-23](2026-10-23-io-mps.md)
- [Disk–GPU loading: HDF5 and QEM, 2026-10-23](2026-10-23-io-loading-limits.md)
- [CUDA IO benchmark, 2026-10-23](2026-10-23-io-cuda.md)
- [Six default acquisition IO comparison, 2026-10-23](2026-10-23-default-six-io.md)
- [Five-acquisition Disk–GPU loading comparison, 2026-10-23](2026-10-23-default-five-io.md)
- [CUDA QEM loading: cached-block scheduling, 2026-10-23](2026-10-23-cuda-qem-loader-fix.md)
- [CUDA QEM loading and file-cache recheck, 2026-10-23](2026-10-23-cuda-qem-cache-recheck.md)