The README mixes feature matrices with benchmark timing tables. The numbers go stale quickly, depend on specific hardware (A6000 GPU, particular CPU), and clutter the document.
Proposal
Remove timing tables and performance claims from README.md. Keep the feature matrices that show which backends each function supports.
Sections to remove
- Read performance tables (real-world files, synthetic tiled)
- Write performance table
- Reproject performance table
- Full pipeline performance table
- Merge performance table
- The "5-7x faster" claim in the GDAL notes section
Sections to keep
- Feature matrices with the backend support markers
- Projection support matrix
- Algorithm descriptions and source citations
Rationale
Benchmark stats fit better in dedicated benchmark docs or release notes where they can be tied to a specific commit and hardware setup. Mixing them into the feature listing makes the README harder to scan and adds maintenance work every time hardware or codecs change.
The README mixes feature matrices with benchmark timing tables. The numbers go stale quickly, depend on specific hardware (A6000 GPU, particular CPU), and clutter the document.
Proposal
Remove timing tables and performance claims from
README.md. Keep the feature matrices that show which backends each function supports.Sections to remove
Sections to keep
Rationale
Benchmark stats fit better in dedicated benchmark docs or release notes where they can be tied to a specific commit and hardware setup. Mixing them into the feature listing makes the README harder to scan and adds maintenance work every time hardware or codecs change.