# gsrf-eval · `gsrf-bench`

**Zero Overshoot (GSRF) — evaluation package (non-production)**  
Company: **BoonMind** · Product: **Zero Overshoot**  
Contact: **info@boonmind.io**

Public evaluation layer only. See **[LICENSE-EVAL.md](./LICENSE-EVAL.md)**.  
Production use requires a commercial license: https://www.zeroovershoot.com/customers.html

## Install (local)

```bash
cd eval
python3 -m pip install -e .
```

Requires Python 3.9+ and NumPy.

## Run `gsrf-bench`

```bash
# Built-in synthetic mid-band + spike (demo)
gsrf-bench --synthetic --normal 50

# Your CSV (column auto-detected: value / signal / y / raw / first numeric)
gsrf-bench --input path/to/trace.csv --normal 50.0

# Named column + custom EMA
gsrf-bench -i trace.csv -c sensor_a --normal 72.5 --ema-alpha 0.1
```

Or without install:

```bash
cd eval
python3 -m gsrf_eval --synthetic --normal 50
```

## What it prints

Side-by-side **raw · EMA · GSRF Practical** for:

| Metric | Meaning |
|--------|---------|
| **osc31** | std of (series − rolling mean 31) — mid-band / high-pass energy |
| **peak_dev** | max \|series − median\| — peak excursion |
| **MAD vs raw** | tracking cost (EMA often wins — **by design** for GSRF spring) |
| **sweet-band spectral power** | rFFT power for periods ≈ 50–120 samples |

Every run starts with an explicit disclaimer: **your data ≠ locked website packs**.

## Library API

```python
from gsrf_eval import run_gsrf, run_ema, osc31, peak_dev, mad_vs_raw

raw = ...  # 1-D positive float array
out = run_gsrf(raw)           # x* = median log by default
ema = run_ema(raw, alpha=0.12)
print(osc31(out), peak_dev(out), mad_vs_raw(out, raw))
```

Defaults match the public `/try` balanced-style Practical identity.

## What this is not

- Not a production binary or validated domain preset pack  
- Not an open-source production grant (MIT/AGPL dual-license is **not** offered here)  
- Not a claim that local CSV results equal excellence hunt / phase6.2 numbers  

## Links

- Site: https://www.zeroovershoot.com/  
- Bench page: https://www.zeroovershoot.com/gsrf-bench.html  
- Evidence: https://www.zeroovershoot.com/evidence.html  
- Try snippet: https://www.zeroovershoot.com/try.html