# Audio-transform inspection — `I am the method`

## Fixed input and views

The input is the 48 kHz stereo PCM derivative of the preserved YouTube MP4,
already identified by the broader video audit.  `scripts/render_video_audio_scales.py`
uses a fixed 8,192-sample Hann window and 2,048-sample hop to render:

1. mid audio `(L+R)/2`, linear frequency;
2. mid audio `(L+R)/2`, logarithmic frequency;
3. side audio `(L-R)/2`, linear frequency;
4. side audio `(L-R)/2`, logarithmic frequency;
5. mid and side residuals after subtracting each frequency bin's temporal
   median; and
6. a fixed-range mid-minus-side spectral-energy view.

The two PNGs are deliberately visual aids, not an extraction algorithm.  Each
channel pair uses the same fixed colour range, and the residual views use a
fixed +/-18 dB range.  This prevents contrast changes from being reported as
a hidden graphic.

## Result

The transforms retain the named musical material, the two internal silent
intervals near 122–125 seconds, and final silence after about 143 seconds.
They do not show a legible textual image, barcode, stable FSK/RTTY trace, or a
channel-exclusive geometric payload in this lossy YouTube audio copy.

This is a bounded negative result.  It does not rule out a transformation not
specified here, phase-only encoding, or information present only in an
unavailable pre-YouTube original.  A claimed next result must name its exact
input, transform, and a control that would not yield the same pattern from an
ordinary rendering of *Contrapunctus I*.

## Reproduce

```powershell
$env:PYTHONPATH = 'tmp\video-tools'
& '<bundled-python>' scripts\render_video_audio_scales.py `
  'solution\video\i-am-the-method-kQx2G3x7_WQ\analysis\audio-48khz-stereo.wav' `
  --output 'solution\video\i-am-the-method-kQx2G3x7_WQ\analysis'
```