measured head-to-head - structured data classes - byte-exact SHA-256 verified
All numbers measured on this server, roundtrip-verified SHA-256 byte-exact. Proofs available on request.
| Input | xz -9 -e | Pedulli (best-of-N) | Δ |
|---|---|---|---|
| MP4 master 10 MB | 9,468,380 B | 9,430,108 B | -38,272 B |
| 1 MiB of zeros | 156 B | 13 B | much smaller (redundant input) |
| 1 MiB periodic binary | 1,108 B | 436 B | -61% |
| JSON 31 KB | 4,360 B | 1,265 B | -71% |
| Long English text 500 KB | ~145 KB | ~145 KB | tie - selects xz's LZMA2 result |
| HTML 161 KB | 35,705 B | 35,705 B | tie - selects xz's LZMA2 result |
| Apache logs 3.5 MB | 211 KB | 211 KB | tie - selects xz's LZMA2 result |
| Random bytes 1 MB | 1,049,000 B (+424 B) | 1,048,577 B (+1 B) | 423 B less overhead |
Switch to Pedulli for video masters, JSON/XML telemetry, periodic sensor streams, and the per-file +1 byte never-worse floor. On long English-text archives Pedulli ties xz by selecting its LZMA2 result - so you are never worse off than xz, and you win on every structured class.