mdziper 0.1.0

Compress Markdown to reduce word count while preserving content

mdziper

Compress Markdown source text to reduce word count while preserving content. Designed for academic peer review rebuttals with strict word limits.

Install

pip install mdziper

Usage

Python API

from mdziper import compress

text = open("rebuttal.md").read()

# Standard mode: rendered output is pixel-perfect identical
result = compress(text, mode="standard")

# Extreme mode: aggressive compression, may alter formatting
result = compress(text, mode="extreme")

print(result.text)
print(f"{result.original_words} → {result.compressed_words} words ({result.savings:.0%} saved)")

CLI

# Standard compression
mdziper input.md -o output.md

# Extreme compression with stats
mdziper input.md --extreme --stats

# Pipe from stdin
cat input.md | mdziper --extreme > output.md

# Skip specific rules
mdziper input.md --extreme --exclude E04 E16

# List all rules
mdziper --list-rules

Compression Modes

Standard Mode

Removes whitespace that doesn't affect Markdown rendering:

• Trailing whitespace on lines
• Extra blank lines
• Spaces inside table cells: | cell | → |cell|
• Spaces in LaTeX math: $a + b = c$ → $a+b=c$
• Spaces in LaTeX braces: $\frac{ a }{ b }$ → $\frac{a}{b}$
• Redundant spaces in link syntax
• Repeated inline links → reference-style links

Extreme Mode

All standard optimizations plus aggressive space removal. Markdown structure (headings, lists, code blocks, tables) is preserved — only cosmetic spaces are removed.

Punctuation spacing:

• word, word → word,word
• Note: text → Note:text
• word; word → word;word
• sentence. Next → sentence.Next
• word (detail) → word(detail)
• and / or → and/or
• word -- word → word--word

Headings:

• Numbered prefix: ## 1. Setup → ## 1.Setup
• Content underscores: ## Experimental Setup → ## Experimental_Setup

Tables:

• Cell content hyphens: |Base A| → |Base-A|

References:

• [1]: Smith et al., "Deep Learning for X", NeurIPS 2023 → [1]:Smith_et_al.,Deep_Learning_for_X,NeurIPS_2023

Common phrases:

• Reviewer 1 → Reviewer-1, Reviewer A → Reviewer-A

Note: Extreme mode produces minor visual differences but does not break Markdown rendering. Headings, lists, tables, and code blocks all render correctly.

How It Works

1. Segment: Split input into protected regions (code blocks, inline code) and compressible regions (text, math)
2. Compress: Apply mode-specific rules to each segment type
3. Reassemble: Join segments and apply global optimizations

Code blocks and inline code are never modified.

License

MIT