Primary credibility path · free · ~10 minutes on your CSV
gsrf-bench
Local CLI: run GSRF Practical vs EMA on your CSV. Self-verified metrics (osc31 · peak_dev · MAD · spectral) beat reading our 52 datasets for a purchase decision. No sales call. No third-party peer review required to start.
Non-production evaluation only.
Full text: LICENSE-EVAL.md.
Your results ≠ website packs (those are still downloadable for click-through).
Production needs a commercial license — gsrf@boonmind.io after you like your numbers.
Install & run
# from the site repo (or GitHub CBoon99/gsrf-papers) cd eval python3 -m pip install -e . # if editable install fails on old pip: # PYTHONPATH=. python3 -m gsrf_eval --synthetic --normal 50 gsrf-bench --synthetic --normal 50 gsrf-bench --input path/to/trace.csv --normal 50.0 gsrf-bench -i trace.csv -c sensor_a --normal 72.5
Package path on this site: /eval/ · Source: github.com/CBoon99/gsrf-papers/tree/main/eval · citations
Metrics (honest side-by-side)
| Metric | What it tells you |
|---|---|
| osc31 | High-pass energy (rolling-31 residual std). Lower = calmer ring. |
| peak_dev | Max |series − median|. Peak pull toward normal. |
| MAD vs raw | Tracking cost. EMA often wins — expected with spring on. |
| Sweet-band spectral power | rFFT power for periods ≈ 50–120 samples. |
Every run prints a terminal banner: local data ≠ phase6.2 / excellence hunt locked packs.
Architecture (eval vs production)
Public evaluation layer
- Website snippet + this package
- Non-production eval license
gsrf-benchfor local CSV proof
Commercial production layer
- Validated domain presets
- Production binary / wheel (single ICP path)
- Tiered license — pricing
What we are not shipping (yet)
- No MIT/AGPL dual-license production core
- No polyglot ports (ROS / Telegraf / …) in this release
- No free-form browser WASM playground
Phase 1 is evidence distribution. Phase 3 is one deep production core for the primary ICP.