pushk

PUSHK Compression Benchmark

This repository contains benchmark results for compressing a structured data file using multiple popular compression algorithms and the experimental PUSHK pipeline.

---

📊 Benchmark: backobjs.cells

File	Size (bytes)	% of original	Compression (%)	Δ vs best (bytes)
backobjs.cells	16384	100.00%	0.00%	+14900
backobjs.cells.bz2	3688	22.51%	77.49%	+2204
backobjs.cells.arj	3419	20.87%	79.13%	+1935
backobjs.cells.zip	3350	20.45%	79.55%	+1866
backobjs.cells.gz	3205	19.56%	80.44%	+1721
backobjs.cells.rar	2323	14.18%	85.82%	+839
backobjs.cells.xz	1848	11.28%	88.72%	+364
backobjs.cells.lzma	1803	11.00%	89.00%	+319
backobjs.cells.7z	1569	9.58%	90.42%	+85
backobjs.cells.pushk	1484	9.06%	90.94%	0 (best)

---

📌 Notes

• Original file size: 16384 bytes
• Best result: PUSHK

• PUSHK compressed data:

  • 1458 bytes (pure compressed stream)
  • +26 bytes container header

• Final size:

  • 1484 bytes total

Header includes:

• File ID
• Format version
• Data type
• CRC checksum
• Other minimal metadata

---

🧠 About PUSHK

PUSHK is not a traditional compressor in the classical sense.

Instead of directly encoding the input stream, it applies a preprocessing pipeline that:

• Reorders data blocks
• Clusters structures by entropy and similarity
• Extracts repeating patterns (tile-aware)
• Reduces entropy before final encoding

Final stage uses standard techniques (e.g. RLE + Huffman), but on transformed data.

---

⚖️ Key Insight

PUSHK does not just compress data — it makes data more compressible.

This allows it to outperform general-purpose compressors on structured inputs.

---

⚠️ Disclaimer

• Results are data-dependent
• PUSHK is optimized for structured / tile-like data

• Performance may vary on:

  • random data
  • already compressed files
  • high-entropy inputs