Decoder comparison
The decoder matrix decodes byte-identical OpenZL frames on both implementations. Each result preserves the same content type, width and string boundaries as its input. The full report includes per-file throughput, ratio and charts.
Implementations and API mode
Reuse prepared decoder contexts and caller output storage. Frame generation, corpus loading, validation and output checks remain outside timing. The published cases use the same level-6, format-27 frames as the encoder matrix; a decoder comparison must record how the compressed corpus was generated.
Interpret the result
Use total uncompressed bytes divided by summed median time for aggregate throughput. Distinguish serial, numeric and record views and their selected files. Backend availability and compiler flags belong to the recorded host, not a universal guarantee. Stored frames, typed node execution and compressed formats have different costs; a short smoke run is not a regression threshold.