# Reference signals bench

Twelve independent, deterministic examples cover ultrasonic and long-period pressure signals, EMF, VOC/eCO2, PM2.5, lightning, body-frame IMU, EEG, skin resistance, pulse oximetry, pupil diameter and a bitstream. None are partner captures or human-subject data.

Sources are bounded adaptations of the sourcebook's larger proposed captures. Ultrasound is 4096 float32 samples at 384 kHz (10.67 ms), plus its complete one-sided DFT. EEG is four seconds at 128 Hz; IMU is eight seconds at 32 Hz; slow pressure spans ten minutes. Other traces span 60 seconds at 1 Hz. The bitstream is 256 deterministic uint8 bits. File layouts, channel units, durations and limitations are explicit in `index.json` and `methods.json`.

## Reproduce

Use Node 22 and a built canonical disclosureos checkout at `88bcbfcc8b17af30837313aff596cec8e1bde43a`:

```sh
node build.mjs /path/to/disclosureos ./rebuilt
```

The unchanged observation, acquisition, binding, processing and product-source-selection contracts validate these artifacts. The spectrum retains its waveform input and method. Canonical selectors identify every CSV row and binary sample. Hash verification covers all displayed sources. Structural validation is not physical calibration or scientific validation; binding context remains explicitly incomplete.

Each device has its own local clock. Selecting a sample changes the local source coordinate, not the Meridian field cursor. Coincident authored origins do not imply synchronization or correlation. Gaps remain gaps. Lightning distance has no bearing. IMU acceleration is not integrated into a trajectory. eCO2 is not measured CO2. Skin resistance is not conductance. Pupil drawings are numerical schematics. Bit counts do not test randomness. Native WAV, MiniSEED, EDF/BDF, compressed logs and camera formats are not decoded here.
