The scoreboard
Verified on 65 real-world files. Every file decompresses byte-exact. Zero losses. Ever.
0
WIN
0
TIE
0
LOSS
CSV
23WIN2TIE
Up to 42.5% smaller than LZMA-9
Best: 42.5%Avg: 18.3%
JSONL
10WIN2TIE
Up to 88.8% smaller than LZMA-9
Best: 88.8%Avg: 43.2%
Logs
1WIN
31.7% smaller on Apache access logs
Best: 31.7%Avg: 23.6%
JSON
1WIN
8.7% smaller on structured JSON
Best: 8.7%Avg: 8.5%
Text
6WIN
14% smaller on literary classics
Best: 14%Avg: 10.9%
Source Code
14WIN6TIE
Even code compresses better
Best: 3.7%Avg: 2.1%
All 65 files
File | Type | Size | Saving vs LZMA | Result |
|---|---|---|---|---|
| stress_10k.jsonl | JSONL | 2.4 MB | 88.8% | WIN |
| sensor_data_20k.jsonl | JSONL | 1.8 MB | 45.4% | WIN |
| user_records_10k.jsonl | JSONL | 1.6 MB | 45.0% | WIN |
| event_log_5k.jsonl | JSONL | 0.9 MB | 44.1% | WIN |
| us_baby_names.csv | CSV | 7.1 MB | 42.5% | WIN |
| consumer_complaints_26k.csv | CSV | 4.5 MB | 39.2% | WIN |
| api_logs_8k.jsonl | JSONL | 1.4 MB | 38.3% | WIN |
| ecb_exchange_rates.csv | CSV | 6.6 MB | 37.2% | WIN |
| jhu_covid_deaths_us.csv | CSV | 11.3 MB | 33.2% | WIN |
| apache_access_10k.log | Log | 1.3 MB | 31.7% | WIN |
| noaa_weather_daily.csv | CSV | 3.2 MB | 28.8% | WIN |
| wide_schema.jsonl | JSONL | 0.3 MB | 27.9% | WIN |
| owid_energy.csv | CSV | 5.8 MB | 26.6% | WIN |
| owid_co2.csv | CSV | 8.4 MB | 25.9% | WIN |
| mixed_types.jsonl | JSONL | 0.5 MB | 24.7% | WIN |
| fivethirtyeight_nba.csv | CSV | 2.1 MB | 24.2% | WIN |
| airport_runways.csv | CSV | 1.8 MB | 22.7% | WIN |
| ecommerce_events.jsonl | JSONL | 1.1 MB | 21.8% | WIN |
| usgs_earthquakes_30d.csv | CSV | 4.2 MB | 21.1% | WIN |
| server_metrics.jsonl | JSONL | 0.8 MB | 18.6% | WIN |
| owid_covid_full.csv | CSV | 95.0 MB | 16.4% | WIN |
| fivethirtyeight_congress.csv | CSV | 1.4 MB | 15.6% | WIN |
| clickstream.jsonl | JSONL | 0.6 MB | 15.3% | WIN |
| airport_frequencies.csv | CSV | 0.9 MB | 14.3% | WIN |
| pride_and_prejudice.txt | Text | 0.75 MB | 14.0% | WIN |
| world_cities.csv | CSV | 2.8 MB | 13.5% | WIN |
| global_power_plants.csv | CSV | 6.3 MB | 12.2% | WIN |
| sherlock_holmes.txt | Text | 0.59 MB | 12.0% | WIN |
| tale_of_two_cities.txt | Text | 0.79 MB | 11.8% | WIN |
| frankenstein.txt | Text | 0.44 MB | 11.6% | WIN |
| un_population.csv | CSV | 3.5 MB | 10.9% | WIN |
| country_indicators.csv | CSV | 1.1 MB | 9.7% | WIN |
| alice_in_wonderland.txt | Text | 0.17 MB | 9.3% | WIN |
| iris.csv | CSV | 0.004 MB | 8.8% | WIN |
| api_config.json | JSON | 0.07 MB | 8.7% | WIN |
| chicago_crimes.csv | CSV | 42.1 MB | 7.9% | WIN |
| titanic.csv | CSV | 0.06 MB | 7.6% | WIN |
| moby_dick.txt | Text | 1.2 MB | 6.9% | WIN |
| boston_housing.csv | CSV | 0.04 MB | 6.5% | WIN |
| wine_quality.csv | CSV | 0.26 MB | 5.3% | WIN |
| auto_mpg.csv | CSV | 0.02 MB | 4.7% | WIN |
| react_index.js | Source | 0.85 MB | 3.7% | WIN |
| lodash.js | Source | 0.54 MB | 3.4% | WIN |
| express_app.ts | Source | 0.12 MB | 3.2% | WIN |
| django_models.py | Source | 0.34 MB | 3.0% | WIN |
| spring_boot.java | Source | 0.67 MB | 2.8% | WIN |
| tensorflow_ops.cc | Source | 0.91 MB | 2.6% | WIN |
| linux_kernel.c | Source | 1.4 MB | 2.5% | WIN |
| rust_compiler.rs | Source | 0.45 MB | 2.4% | WIN |
| go_stdlib.go | Source | 0.38 MB | 2.2% | WIN |
| swift_foundation.swift | Source | 0.29 MB | 2.1% | WIN |
| vue_runtime.js | Source | 0.72 MB | 2.0% | WIN |
| angular_core.ts | Source | 0.63 MB | 1.9% | WIN |
| numpy_core.py | Source | 0.21 MB | 1.8% | WIN |
| rails_active_record.rb | Source | 0.48 MB | 1.7% | WIN |
| ghcnd_stations.txt | CSV | 9.1 MB | 0% | TIE |
| movietweetings_ratings.dat | CSV | 0.5 MB | 0% | TIE |
| github_archive_10mb.jsonl | JSONL | 10.0 MB | 0% | TIE |
| nested_objects.jsonl | JSONL | 0.007 MB | 0% | TIE |
| small_util.js | Source | 0.003 MB | 0% | TIE |
| config.yaml | Source | 0.002 MB | 0% | TIE |
| readme.md | Source | 0.004 MB | 0% | TIE |
| setup.cfg | Source | 0.001 MB | 0% | TIE |
| makefile | Source | 0.002 MB | 0% | TIE |
| dockerfile | Source | 0.001 MB | 0% | TIE |
65 files shown — All benchmarked against LZMA-9 (maximum compression)
Methodology
Baseline: LZMA-9 (maximum compression level)
Verification: Byte-exact round-trip on every file
Guarantee: Never-worse (PZIP <= LZMA always)
Tests: 916 automated, zero manual overrides
Container: PZ01 v2 (21-byte overhead)
Date: 2026-02-01About that Weissman Score...
Remember the Weissman Score from Silicon Valley? It measures compression ratio times the log of speed. The show made it up for TV drama.
We made it real. Except ours is verified on 65 real-world files, not fictional middleware. Unlike Pied Piper, we have 46 papers. They had...enthusiasm.
"This guy compresses." — probably Russ Hanneman
Download the test data
Every benchmark above uses public data. Download the same files, run any compressor you want, and verify our numbers yourself. No trust required — just math.
US Baby Names
US Census Bureau — CSV
ECB Exchange Rates
European Central Bank — CSV
Chicago Crimes
City of Chicago — CSV
COVID-19 Data
Johns Hopkins University — CSV
Project Gutenberg
Project Gutenberg — Text
OWID Energy/CO2
Our World in Data — CSV
NOAA Weather
NOAA — CSV
USGS Earthquakes
USGS — CSV
Want to test on your own data? Head to the demo page and upload any file. We'll compress it with PZIP, and you can run LZMA, gzip, bzip2, zstd, and xz right there to compare.