Building a compressor that competes on complete archive size

Compression experiments can look compelling when they report only the transformed file or one favorable sample. I’m building PIZD, a Rust project exploring pattern discovery and structural coding, and I want its comparisons to answer a more practical question: what happens to the complete archive?

Start with a reversible contract

Every candidate has to preserve exact restoration. A smaller output is not useful if decoding changes even one source byte.

Measure the whole result

A transform can add metadata. The relevant comparison includes that overhead and the rest of the archive, not just the part the transform targets.

Make trade-offs visible

Some experimental formats need a current reader, and extra analysis can cost time. Those are part of the result, not footnotes to hide.

PIZD is still an unpublished release candidate. I’m sharing the methodology and would welcome feedback on what workloads and baselines should be included next.

Project and measurements: https://pizd.spedo.dev/index.html