ARCANADA
BUILDING L3 · target L5

Cubrim

Self-improving lossless compressor — autonomous agents race compression hypotheses.

A Rust lossless compression engine built on an N-dimensional cube model with φ-mixed-radix decomposition, a distance map, and layered BWT, context-mixing, and columnar coding. Beats gzip across the full test corpus and approaches zstd-19 on logs and telemetry. The only project in the ecosystem where autonomous agents propose, test, and benchmark compression hypotheses end-to-end — the operator reviews and merges winning branches. Private repo; results published at cubrim.com.

Capabilities

  • N-dimensional cube model — lossless compression on any binary input
  • φ-mixed-radix decomposition + distance map pipeline
  • BWT, context-mixing, and columnar coding layers
  • Beats gzip everywhere; approaches zstd-19 on logs and telemetry
  • Public hypothesis race at /evolution — real numbers, dead ends shown
  • Autonomous agent loop: hypothesis → implementation → benchmark → merge gate

Current autonomy level

L3
What levels mean →

Weakest link

Agents self-grade benchmarks against gzip/zstd; no independent third-party corpus yet. No iteration ceiling — an agent could exhaust compute on a dead-end hypothesis without a hard budget cap.

Roadmap to L5

  1. L3 → L4 — add hard iteration cap per hypothesis run; wire benchmark result to Scrutator index for cross-hypothesis retrieval.
  2. L4 → L5 — full autonomy: agents propose, implement, benchmark, and select winning hypotheses without human merge gate; operator receives summary digest only.

Links

Next step

Review the current build and its development plan

BUILDING means active development, not a promise of public availability. Use the current capabilities and roadmap above as the reference.

Review the path to the target level