LowPack 0.1.2 — The personal documentation release

I built LowPack as a personal project, but parts of the documentation had
started to sound like they came from a product committee. v0.1.2 is me bringing
the explanation back to the original voice: clear about what I was thinking,
honest about the alpha edges, and still precise where the bytes and security
rules matter.

What changed

The README now starts with the simple idea behind LowPack: prepare data for how
it will be stored and used, then rely on proven lossless codecs. I also rewrote
the surrounding guidance so profiles, determinism, security, benchmarking, and
limitations explain why I made each choice instead of only listing behavior.

Oh! The review caught one factual mismatch too. The limitations page said
source dictionary training was inactive, but LowPack already performs bounded,
deterministic training and authenticates the dictionary in the manifest. That
reference is now accurate.

The exact binary tables, safety limits, benchmark values, and compatibility
warnings remain technical on purpose. Personal does not mean vague.

A small guardrail for later

I added a documentation-integrity test that keeps the package version,
changelog, local Markdown links, and README cover image aligned. It is a small
test, but it protects the details readers notice first.

Validation

• 32 tests passed; one Windows symlink test skipped where symlink privileges
  are unavailable.
• Ruff and strict Mypy passed.
• Pull-request and merged-main matrices passed on Python 3.9–3.14.
• Every local Markdown link and the cover asset passed the new integrity test.
• Wheel and source distribution reproduced byte-for-byte.
• A clean wheel installation completed pack, full verify, and unpack.

Install

python -m pip install lowpack-0.1.2-py3-none-any.whl
lowpack --version

This is still an experimental alpha. I would keep the original data, verify
untrusted archives fully, and read the security guidance before extraction.