CPB — Cocoa Powder Bottle

What is CPB?

A multi-layer archive format that compresses, protects, and disguises your data.

📖

Structure Dict

L3 Genre DSL — auto-detects data type, replaces patterns with compact references.

🔤

String Dict

L5 GenDict — caches exact matches. 2nd pass compresses to ~270 bytes.

🛡️

Error Recovery

Reed-Solomon codes. Auto-repair corruption without re-download.

🎬

Video Container

Output as AVI / MP4 / ZIP / PDF / PNG. Standard tools work natively.

🔀

4D–16D Shuffle

Multi-dimensional byte rearrangement. Up to 16 independent axes.

🔍

Full-Text Search

FIDX index — search inside archives without extraction.

CPB vs Existing Tools

64 KB JSON — single-threaded benchmark

gzip

ratio: 34%
enc: 150 MB/s
dec: 400 MB/s

baseline

zstd

ratio: 28%
enc: 380 MB/s
dec: 1,100 MB/s

fastest modern

LZMA/7z

ratio: 22%
enc: 18 MB/s
dec: 120 MB/s

best ratio

zlib

ratio: 38%
enc: 140 MB/s
dec: 350 MB/s

ubiquitous

CPB STD
ratio: 0.8%
enc: 4.2 MB/s
dec: 891 MB/s
learn+dict

CPB L5×2
ratio: 0.4%
enc: 147 MB/s
dec: 1,857 MB/s
cache hit!

Compression Ratio — lower is better (JSON 64 KB)

gzip

34 %

zlib

38 %

LZMA/7z

22 %

zstd

28 %

CPB STANDARD

0.8 %

CPB L5×2

0.4 %

Real-World Example

2,000-entry JSON access log — 401 KB input, round-trip verified ✓

{ "timestamp": "2026-01-01T00:00:00Z", "level": "INFO", "service": "api", "request_id": "req-100000", "user_id": 5506, "path": "/api/v1/orders", "status": 200, "latency_ms": 72, "message": "Request processed successfully" } // × 2,000 entries

Method	Output	Ratio	Notes
gzip (≈L2)	65 KB	15.9%	baseline
CPB STANDARD	74 KB	18.2%	full pipeline, no dict
CPB + L3 dict	65 KB	15.9%	phrase dict on same data
CPB LEARN pass 1	65 KB	15.9%	first time — cache built
CPB LEARN pass 2	270 B	0.07%	cache hit

When CPB wins

Repeated log archival — L5 cache → near-zero size

Structured telemetry — L3 dict kills repeated keys

Sensitive data — RS + L4 shuffle for protection

When CPB loses

Random / encrypted data — no redundancy (~115%)

One-shot compression — L5 cache cold, no edge

Tiny files (<1 KB) — header overhead dominates

Pipeline

Up to 6 layers. Each independently toggleable. 7 preset profiles.

📁

Input Data

Any file or folder

↓

📖

L5 GenDict

Exact-match cache — 0.07% on 2nd pass

↓

🔤

L3 Genre DSL

Structure-aware phrase replacement

↓

📦

L2 Compress

LZMS / MSZIP / XPRESS — 0.6% ratio

↓

🛡️

L1 Protect

Reed-Solomon error correction

↓

🔀

L4 Shuffle

4D–16D multi-dimensional rearrangement

↓

🎬

Container

.cpb / .zip / .mp4 / .pdf / .png

Detailed Benchmarks

Linux x86-64 / g++ 13 / -O2 / C++17 / single-threaded / best of 5 runs

L2 Algorithm — Encode MB/s

AUTO

631 MB/s

LZMS

895 MB/s

MSZIP

895 MB/s

XPRESS_HUFF

892 MB/s

XPRESS

892 MB/s

RLE

240 MB/s

DELTA

293 MB/s

NONE

12,435 MB/s

L1 RS Protection — Encode MB/s

STANDARD

17.6 MB/s

MAX

4.5 MB/s

STEGO

1.1 MB/s

LIGHT

17.7 MB/s

NONE

14,282 MB/s

L4 Dimensions — Encode Speed

24 MB/s

12.3 MB/s

12D

8.1 MB/s

16D

6.1 MB/s

Pipeline — Compression Ratio (64 KB JSON)

Configuration	Ratio	Note
L2 only	0.6%	fastest
L3 + L2	0.8%
STANDARD (4D)	0.8%	default
STANDARD + L3	1.2%
STANDARD + 16D	0.8%
STANDARD + L3 + 16D	1.2%
ARCHIVE	114.4%	expands (protection only)

Data Type Sensitivity (STANDARD pipeline)

Data	Ratio
JSON 1 KB	26.4%
JSON 64 KB	0.8%
JSON 1 MB	0.7%
Random 64 KB	115.2%
Random 1 MB	114.8%
All-zeros 1 MB	0.0%

L5 LEARN Cache + Carrier Overhead

L5 Pass	Ratio	Decode MB/s
pass 1 (cold)	1.2%	855
pass 2 (hit)	0.4%	1,857
pass 3	0.4%	2,531

Carrier	Overhead
CPB (.cpb)	+0 B
MP4 (.mp4)	+56 B
PNG (.png)	+93 B
ZIP (.zip)	+224 B
PDF (.pdf)	+883 B

Roadmap

v0.1 — Core Pipeline

L2 compression, L1 Reed-Solomon, L4 shuffle, 5 carrier formats. 654 tests.

v0.2 — Dictionary System

L3 Genre DSL, L5 GenDict cache, dictionary training, FIDX search.

v1.0 — Public Release

Open source, documentation, benchmarks, GitHub Pages.

v1.1 — SDK

Embeddable library API. Client-to-Client data foundation.

v2.0 — Advanced

Hilbert curve, differential backup, multi-frame random access.