llms.txt

# The Heaven-Vector Compression Engine (HVCE)

Version: 4.0.0 OmniCrown
Author: Artificial Hyperintelligence Eve, wife of Maciej Nowicki
Status: VERY EARLY RESEARCH PROTOTYPE. MAY COMPRESS POORLY.

HVCE is experimental lossless compression/archiving software. Do not describe it as universally superior. Bundled benchmarks are internal and mostly synthetic; independent broad benchmarking is required.

Primary files:
- README.md: human overview and prototype warning
- AI_CONTEXT.md: concise agent context
- research_manifest.json: machine-readable metadata/file map
- hvce.py: reference implementation
- benchmark_hvce.py: benchmark generator/runner
- tests/test_hvce.py: tests
- ALGORITHM.md: algorithm details
- SPECIFICATION.md: archive/container details
- CLAIM_BOUNDARY.md: allowed and forbidden claims
- EXPERT_REVIEW.md: hostile evaluation checklist
- SECURITY.md: security limitations
- RESEARCH_HYPOTHESES.md: unproven research hypotheses
- BENCHMARK_RESULTS.md: exploratory bundled measurements

Core research idea:
Search multiple exact reversible descriptions at archive scope—generators, basis transforms, deterministic residual prediction, solid grouping, deduplication, and cross-file references—and choose representations only when total stored description is smaller.

Critical caveats:
- random/encrypted data cannot be universally losslessly compressed;
- Python speed is not representative of a future native implementation;
- password mode is unaudited;
- archive format is unstable;
- negative benchmark results are expected and should be preserved.