pushk

PUSHK Compression Benchmark

This repository contains benchmark results for compressing a structured binary data file using multiple popular compression algorithms and the experimental PUSHK pipeline in Resolution Drop (RDrop) mode.

📊 Benchmark: `example7b_rdrop.bin`

⚠️ Disclaimer

Results are highly data-dependent
Resolution Drop mode is heavily optimized for structured / tile-like / patterned / gradient-heavy binary data
Performance may vary significantly on random data, high-entropy inputs, or already compressed files

File	Size (bytes)	% of original	Compression (%)	Δ vs best (bytes)
example7b_rdrop.bin	399	100.00%	0.00%	+72
example7b_rdrop.bin.bz2	493	123.56%	-23.56%	+166
example7b_rdrop.bin.arj	517	129.57%	-29.57%	+190
example7b_rdrop.bin.zip	528	132.33%	-32.33%	+201
example7b_rdrop.bin.gz	378	94.74%	5.26%	+51
example7b_rdrop.bin.rar	457	114.54%	-14.54%	+130
example7b_rdrop.bin.xz	372	93.23%	6.77%	+45
example7b_rdrop.bin.lzma	327	81.95%	18.05%	0
example7b_rdrop.bin.7z	462	115.79%	-15.79%	+135
example7b_rdrop.bin.pushk	333	83.46%	16.54%	+6

% of original = (size / original size) * 100
Compression (%) = 100 - % of original
Δ vs best = size - pushk size

📌 Notes

Original file: example7b_rdrop.bin
Mode: test_run_02_rdrop — Resolution Drop (RDrop)
PUSHK compressed data:
- 317 bytes — pure compressed stream (final RDropblock)
- +16 bytes container header (stPushFileHeader)
Final size:
- 333 bytes total

Offset	Size	Field	Description
0	4	`hdr`	File ID — `'PSHK'`
4	1	`mver`	Module version
5	1	`sver`	Stream/compression version
6	1	`opt1`	Option flags 1
7	1	`opt2`	Option flags 2
8	4	`fsize`	Original file size (bytes)
12	4	`crc`	CRC32 checksum of original data

Header includes:

File ID
Format version
Data type
CRC checksum
RDrop metadata (layers, key count, saturation stats)
Other minimal metadata

🧠 About PUSHK — Resolution Drop (RDrop) Mode

PUSHK is not a traditional compressor in the classical sense.

In RDrop mode, it applies a sophisticated preprocessing pipeline that:

Iteratively scans data and saturates repeating patterns
Builds a compact Global Key Database (55 keys in this run)
Applies hierarchical Resolution Drop across multiple layers, discarding low-value resolution information
Dramatically reduces entropy on highly structured data before final encoding

This run specifically demonstrates the power of the RDrop technique on tile-like / gradient-heavy / packed binary data.

📂 Generated Files from This Run

Logs generated by PUSHK pre-filter:

example7b_rdrop.bin.pushk_log_cells.txt — Iterative cell saturation process (55 iterations)
example7b_rdrop.bin.pushk_log_gkeys.txt — Complete Global Key Database (55 keys with rates and types)
example7b_rdrop.bin.pushk_log_reso.txt — Detailed Resolution Drop layers and statistics (final 399-byte block)

Visualizations:

example7b_rdrop.bin_scn.svg — Full visualization of scanning, pattern detection
spctr_B0_spc.svg — Huffman tree for block #0
spctr_B1_spc.svg — Huffman tree for block #1
spctr_B2_spc.svg — Huffman tree for block #2
spctr_B3_spc.svg — Huffman tree for block #3
spctr_B4_spc.svg — Huffman tree for block #4
spctr_B5_spc.svg — Huffman tree for block #5
spctr_B6_spc.svg — Huffman tree for block #6

Optional:

spctr_declined_global_no_localization_352_bytes_result.svg
Global summary visualization using a single global Huffman tree — resulted in 352 bytes.
This approach was rejected in favor of multiple smaller, localized trees, which achieved 333 bytes total, providing better compression efficiency.

⚖️ Key Insight

PUSHK does not just compress data — it deeply understands the internal structure and transforms it into a far more compressible representation.

The combination of Global Keys + Hierarchical Resolution Drop allows it to outperform general-purpose compressors on data with strong internal regularity.