hestiaos-benchmarks

Reproducible HestiaOS test & benchmark harnesses — for validation, evidence generation, and external / playground evaluation preparation.

This repo is not an edition repo and not a runtime deployment. Benchmarks generate evidence artifacts; they do not modify external systems.

Role

- reproducible tests
- benchmarks
- validation harnesses
- evidence generation
- playground / external-validation preparation

Contents

benchmarks/
  jupiter-dgs-stage0-stage1/   First benchmark: portable JUPITER/JSC playground prep —
                               Stage 0 environment probe + Stage 1 mock A/B DGS benchmark
                               (governed vs ungated). stdlib-only primary path.

Not to be confused with

science-edition  -> the actual HestiaOS Science Edition (reserved; NOT this repo).

The benchmark here is a harness / portability probe / DGS-vs-baseline evaluation / evidence generator — not a full Science Edition, not a Jülich/JUPITER-validated system or result, not foundation-model training, not a production deployment.

Quick pointer

See benchmarks/jupiter-dgs-stage0-stage1/QUICKSTART.md. Primary path is stdlib-only:

cd benchmarks/jupiter-dgs-stage0-stage1
python3 benchmark/run_stage0_environment_probe.py --out reports
python3 benchmark/run_stage1_dgs_benchmark.py --mode mock --out reports

Safety

No network, no external/destructive actions, no scheduler/HPC jobs, no secrets in reports. See each benchmark's CLAIM_BOUNDARY.md and the repo SECURITY.md / PROVENANCE.md.