26 C++ and python libraries each benchmarked against the current category leader with documented losses. NSSort is 182,000x faster than IPS4o on structured distributions. NSKVCache compresses 1M token context to 168MB from 35GB raw FP16. NSFix runs 18.8x QuickFIX throughput. NSComp decompresses 29x faster than zstd. Every product documents where it loses. Run any benchmark yourself with one make command. Free for individuals. More information at ofmagnitude.com
Every benchmark in this suite documents its loss cases. NSSort loses on structureless data. NSComp loses on random bytes. The wins are real and so are the losses. Clone it and run the benchmarks yourself, one make command per product. Happy to answer questions on methodology.