🦀 MBFA — MidMans Bit Folding Algorithm

MidManStudio | Extended Benchmark Results

Build: #317
Branch: master
Commit: 12fa54d0
Date: Thu Mar 19 18:22:58 UTC 2026
📚 Canterbury Corpus → 📦 Archive Benchmark → 🔬 Algorithm Walkthrough 🎮 Special Benchmark
Compression Results
MBFA
gzip
zstd
xz
brotli
zpaq
C = compress ms   D = decompress ms

💡 Click any row for a detailed breakdown

Dataset Original MBFA gzip zstd xz brotli zpaq RT Visual
About MBFA — Conceptual Status

What MBFA is: MidMans Bit Folding Algorithm is a novel multi-fold iterative compression algorithm under active research and development by MidManStudio. The core concept — encoding data as an instruction chain where each fold produces a bitstream that reconstructs the previous fold — is architecturally distinct from existing published compression algorithms.


Current implementation status: Research implementation with joint Huffman entropy coding on the fold output layer. All roundtrips pass. Active improvements ongoing.


Where MBFA wins: Highly repetitive data — the multi-fold convergence dramatically outperforms gzip and zstd by exploiting structure that single-pass algorithms cannot see across their own metadata.


PNG / JPEG / FBX notes: PNG is already DEFLATE-compressed internally. JPEG at q85 is already DCT-quantised and Huffman-coded. Both are near-incompressible for any general-purpose compressor — all algorithms including gzip and zstd will show ratios close to 100%. FBX binary contains float mesh data (vertices, normals, UVs) which has moderate LZ structure. Results above 95% on these formats are expected and correct behaviour.


Current prose performance: Joint Huffman on the fold output layer brought prose compression to within ~2–3% of gzip on large files. The remaining gap is the flat-cost BACKREF offset field (15 raw bits each) — a known target for further improvement.


Lower % = better compression. 100% = incompressible data correctly passed through with near-zero overhead.