OneCharacterCode benchmark - test run report
===============================================

Started   : 2026-05-11T12:31:13
Finished  : 2026-05-11T12:31:15
Duration  : 1.62 seconds
Machine   : LIMITLESS
PowerShell: 5.1.26100.8115
Brotli    : False

Command run:
  powershell -File run_benchmark.ps1

Inputs tested:
  JSON_AGENT_SAMPLE.json  (5,458 bytes)
  SIMPLE_HTML_SAMPLE.html  (6,904 bytes)
  TEXT_ARTICLE_SAMPLE.txt  (8,824 bytes)

Results table:
  File                                    Raw       Gzip     Brotli        OCC  Recon
  --------------------------------------------------------------------------------------
  JSON_AGENT_SAMPLE.json                5,458      2,529        n/a      6,188   PASS
  SIMPLE_HTML_SAMPLE.html               6,904      2,929        n/a      8,736   PASS
  TEXT_ARTICLE_SAMPLE.txt               8,824      3,529        n/a     10,592   PASS

Compression ratios:
  JSON_AGENT_SAMPLE.json            gzip=53.66%  brotli=n/a  occ=-13.37%
  SIMPLE_HTML_SAMPLE.html           gzip=57.58%  brotli=n/a  occ=-26.54%
  TEXT_ARTICLE_SAMPLE.txt           gzip=60.01%  brotli=n/a  occ=-20.04%

Reconstruction status (SHA-256 round-trip):
  JSON_AGENT_SAMPLE.json            PASS
  SIMPLE_HTML_SAMPLE.html           PASS
  TEXT_ARTICLE_SAMPLE.txt           PASS

Limitations:
  - OCC column is from a PROTOTYPE symbolic dictionary encoder, not the
    final patented OneCharacterCode engine. The prototype produces a real
    reconstruction round-trip and an honest, reproducible compression ratio.
  - On these inputs, gzip (and Brotli, where available) are mature and
    well-tuned. A simple substring-dictionary prototype often will not beat
    them. The honest result is reported as-is.
  - These three inputs are short (KB-scale). Compression performance on
    larger and more diverse inputs is different; results here should not
    be generalized.
  - Brotli requires .NET Core 2.1+ / PowerShell 7+. On Windows PowerShell
    5.1 / .NET Framework 4.x, the Brotli column reads n/a.

Next benchmark steps:
  - Add KB-to-MB scale inputs (whole websites, archive sets, log corpora).
  - Compare against zstd, xz, lzma in addition to gzip and Brotli.
  - Replace the prototype encoder with the production OCC engine and rerun
    the same harness for an apples-to-apples comparison.
  - Independent reproducibility: a third party runs run_benchmark.ps1 on
    their own machine against the same inputs and compares hashes.

All inputs and outputs are hashed in SHA256_MANIFEST.txt for verification.
Reproducibility instructions: README_REPRODUCE.txt