# ZJX telemetry comparator campaign public source note

Published: 2026-09-05
Campaign: zjx-site-telemetry-comparator-2026-09-05
Started (UTC): 2026-09-06T03:23:08Z · Finished (UTC): 2026-09-06T04:07:30Z
Status: featured public comparison for the ZJX product site

## Claim boundary

On one real service telemetry JSONL capture, ZJX 1.27 CIPHERGLASS on the compact objective wrote a complete archive of 4,412,968 bytes. The strongest of the fourteen world archive baselines, zpaq -m5, wrote 7,125,674 bytes. ZJX is therefore 38.1% smaller than the strongest comparator and 98.7% smaller than the raw input. The fast objective wrote 4,412,968 bytes. Every archive, ZJX and comparator alike, was restored and matched against the source SHA-256.

This is one dataset and one build. It is not a universal compression-ratio claim and does not predict any other workload. Wall time and peak memory were observed in single sequential runs on a development workstation that was concurrently running unrelated work; they are recorded below as observations and are not performance evidence. ZJX performance qualification runs on a dedicated controlled host under its own campaign law.

Product site: <https://zjx.greyforge.tech/>
Brief section: <https://zjx.greyforge.tech/docs.html#results>
Benchmark ledger: <https://greyforge.tech/products/zjx/benchmarks>

## Identity

| Item | Value |
|---|---|
| Input | one JSONL file, 342,409,609 bytes, SHA-256 `9775bf5d9a352960a857401479ba3f314d4ed8d65a4d24b52014b1dee60605ec` |
| Records | 681,648 (schema census of the same capture) |
| Content | private service telemetry captured 2026-06-03; not published |
| ZJX binary | `zjx 0.1.0 (8827fe367000d39e994f4668603544d5c07406d2, ReleaseSafe)` |
| ZJX binary SHA-256 | `1bf650ea0c605cba1b3fef221b1ff03106b473557e94eca9c9785c504267108e` |
| ZJX build | release/zjx-1.26-relicwire lineage, CIPHERGLASS closure commit 8827fe36, ReleaseSafe, pinned compiler toolchain |
| Byte-identity note | ZJX 1.28 LOCKWIRE writes identical plaintext bytes for this input class plus a 16-byte manifest-integrity footer (LOCKWIRE byte-identity campaign) |
| Host | greyforge, Linux 7.0.0-30-generic, x86_64, AMD Ryzen 5 3600 6-Core Processor, 15 GiB RAM |
| Timing authority | NON-AUTHORITATIVE development workstation; sequential single runs; no repetition statistics |
| Tar wrapper | POSIX tar, sorted names, owner/group 0, fixed mtime; 342,415,360 bytes, SHA-256 `0678ac1c0b446fc4303fa6adaaaa82a84a6b3102898f74e463f8230b4e565e4c` |

Tool versions: tar tar (GNU tar) 1.35; bzip2 bzip2, a block-sorting file compressor.  Version 1.0.8, 13-Jul-2019.; xz xz (XZ Utils) 5.4.5; zstd *** Zstandard CLI (64-bit) v1.5.5, by Yann Collet ***; brotli brotli 1.1.0; lz4 *** LZ4 command line interface 64-bits v1.9.4, by Yann Collet ***; zip Copyright (c) 1990-2008 Info-ZIP - Type 'zip "-L"' for software license.; gzip gzip 1.12; 7-Zip 7-Zip 23.01 (x64) : Copyright (c) 1999-2023 Igor Pavlov : 2023-06-20; zpaq zpaq v7.15 journaling archiver, compiled Jan  5 2021; rar RAR 7.00   Copyright (c) 1993-2024 Alexander Roshal   26 Feb 2024; GNU time time (GNU Time) UNKNOWN.

## Method

ZJX archived the file directly with the compact and fast objectives, then the native archive test verified every payload digest and a native restore recovered the file, whose SHA-256 was compared with the source. Stream compressors (bzip2, xz, zstd, brotli, lz4, gzip) compressed the deterministic tar and were verified by decompressing to a stream whose SHA-256 was compared with the tar. Container archivers (zip, 7-Zip, zpaq, rar) archived the file directly and were verified by extracting and comparing the file SHA-256. Comparator arguments are the ones the ZJX laboratory's world baseline set uses. Runs were sequential; each creation was wrapped in GNU time to record wall, CPU, and peak RSS.

## Results

| Archiver and settings | Complete archive bytes | Versus strongest comparator | Restored and matched | Wall s (indicative) | User+sys s | Peak RSS KiB |
|---|---:|---|---|---:|---:|---:|
| ZJX 1.27 compact | 4,412,968 | 38.1% smaller | yes | 255.1 | 253.5 | 668,748 |
| ZJX 1.27 fast | 4,412,968 | 38.1% smaller | yes | 233.7 | 230.5 | 668,764 |
| zpaq -m5 | 7,125,674 | strongest comparator | yes | 151.5 | 764.3 | 3,831,408 |
| zstd --ultra -22 --long=31 (tar) | 7,280,110 | 2.2% larger | yes | 530.6 | 529.9 | 1,144,164 |
| brotli -q 11 (tar) | 7,345,538 | 3.1% larger | yes | 352.4 | 349.6 | 186,748 |
| xz -9e (tar) | 7,612,052 | 6.8% larger | yes | 390.2 | 389.6 | 657,060 |
| zstd -19 --long=31 (tar) | 7,800,328 | 9.5% larger | yes | 178.0 | 177.9 | 556,644 |
| zstd -19 (tar) | 8,149,050 | 14.4% larger | yes | 177.5 | 177.6 | 217,204 |
| 7-Zip LZMA2 -mx=9, single thread | 9,392,518 | 31.8% larger | yes | 81.3 | 81.2 | 693,576 |
| 7-Zip LZMA2 -mx=9 | 9,666,962 | 35.7% larger | yes | 44.5 | 64.7 | 1,539,848 |
| RAR5 -m5 | 9,843,396 | 38.1% larger | yes | 7.7 | 49.2 | 565,756 |
| 7-Zip PPMd -mx=9 | 12,612,270 | 77.0% larger | yes | 16.1 | 16.1 | 269,448 |
| bzip2 -9 (tar) | 34,456,468 | 383.6% larger | yes | 32.0 | 32.0 | 7,604 |
| gzip -9 (tar) | 40,176,031 | 463.8% larger | yes | 5.6 | 5.6 | 1,704 |
| zip -9 (deflate) | 40,176,334 | 463.8% larger | yes | 5.9 | 5.7 | 2,460 |
| lz4 -12 (tar) | 56,999,958 | 699.9% larger | yes | 9.6 | 9.6 | 6,196 |

Comparators completed and verified: 14/14. Complete archive bytes include format overhead.

## Exact commands

- ZJX compact and fast: the native create command with the named objective and JSON output, then the native archive test and restore commands. ZJX command recipes are not published; the evaluation report reproduces them for the customer.
- `bzip2 -9 (tar)`: `bzip2 -9 -c` · restore: `bzip2 -d -c`
- `xz -9e (tar)`: `xz --format=xz --threads=1 --check=crc64 -9e --stdout` · restore: `xz -d -c`
- `zstd -19 (tar)`: `zstd -q --threads=1 -c -19` · restore: `zstd -d -c`
- `zstd -19 --long=31 (tar)`: `zstd -q --threads=1 -c -19 --long=31` · restore: `zstd -d -c --long=31`
- `zstd --ultra -22 --long=31 (tar)`: `zstd -q --threads=1 -c --ultra -22 --long=31` · restore: `zstd -d -c --long=31`
- `brotli -q 11 (tar)`: `brotli -q 11 -c` · restore: `brotli -d -c`
- `lz4 -12 (tar)`: `lz4 -12 -q -c` · restore: `lz4 -d -c`
- `zip -9 (deflate)`: `zip -9 -q -r <work>/out/zip-deflate-9.zip 20260603.jsonl` · restore: `unzip -q -o <work>/out/zip-deflate-9.zip`
- `gzip -9 (tar)`: `gzip -9 -c` · restore: `gzip -d -c`
- `7-Zip LZMA2 -mx=9`: `7z a -t7z -m0=lzma2 -mx=9 -ms=on -bd -bb0 <work>/out/7z-lzma2-9.7z 20260603.jsonl` · restore: `7z x -y -bd -bb0 <work>/out/7z-lzma2-9.7z`
- `7-Zip LZMA2 -mx=9, single thread`: `7z a -t7z -m0=lzma2 -mx=9 -ms=on -mmt=1 -bd -bb0 <work>/out/7z-lzma2-9-1t.7z 20260603.jsonl` · restore: `7z x -y -bd -bb0 <work>/out/7z-lzma2-9-1t.7z`
- `7-Zip PPMd -mx=9`: `7z a -t7z -m0=PPMd -mx=9 -ms=on -mmt=1 -bd -bb0 <work>/out/7z-ppmd-9.7z 20260603.jsonl` · restore: `7z x -y -bd -bb0 <work>/out/7z-ppmd-9.7z`
- `zpaq -m5`: `zpaq add <work>/out/zpaq-m5.zpaq 20260603.jsonl -m5` · restore: `zpaq extract <work>/out/zpaq-m5.zpaq -to <work>/restore/zpaq-m5`
- `RAR5 -m5`: `rar a -ma5 -m5 -idq <work>/out/rar-rar5-m5.rar 20260603.jsonl` · restore: `rar x -idq -y <work>/out/rar-rar5-m5.rar`

## Limitations

- One dataset, one capture, one build. The private content is not published; the identity above lets a future run on the same bytes be recognized, not reproduced by third parties.
- Timings are single runs on a shared development workstation and carry no repetition statistics; they are not release performance evidence.
- The comparator set is the laboratory's fourteen world baselines at strong settings; other tools, settings, or preprocessing (for example columnar conversion) were not tested.
- Verification compared restored bytes with the source. It did not compare filesystem metadata, which the comparators handle differently by design.
- The fast objective is reported for completeness; its qualified creation-time benefit was measured in the NIGHTMARKET campaign on other corpora and is not re-measured here.