# Stereo voice audit — `I am the method`

## Scope

This bounded test asks whether the left and right soundtrack channels prefer complementary subsets of the four voice tracks in the documented Contrapunctus I MIDI source. It is conditional on the already-recorded partial audio-to-score timing alignment; it does not attempt to turn Bach or panning into a cipher.

## Result

The observed channel preference is stronger than the recorded circular time-shift controls. That supports a score-locked stereo preference, but it is still not a textual decode or proof of deliberate encoding.

- Best observed direction: left voices `2+3+4`; right voices `1`.
- Absolute preference: `0.06985`.
- Circular time-shift control p-value: `0.0050` (200 fixed-seed trials).
- Register check: median spectral centroid is `1228.4 Hz` left and `850.5 Hz` right; MIDI voice 1 is the highest-register voice.

## All complementary partitions

| Left voices | Right voices | Signed preference | Absolute preference |
|---|---|---:|---:|
| 1 | 2+3+4 | -0.06985 | 0.06985 |
| 1+3+4 | 2 | 0.03751 | 0.03751 |
| 1+2+4 | 3 | 0.03015 | 0.03015 |
| 1+2 | 3+4 | -0.02827 | 0.02827 |
| 1+3 | 2+4 | -0.02188 | 0.02188 |
| 1+2+3 | 4 | 0.01281 | 0.01281 |
| 1+4 | 2+3 | -0.00818 | 0.00818 |

## Controls and limits

- The control preserves the channels' spectra and their relationship to each other but circularly shifts both against the score. It evaluates the maximum across all partitions on every trial, so the reported p-value includes partition selection.
- MIDI voices are a reference representation, not stems from the performance. The centroid check runs in the opposite direction from a simple “high register panned right” explanation, so no physical mix explanation is inferred from this chroma result alone.
- A surviving effect would call for comparison with an independent performance/source before any narrative claim; a null result only closes this simple L/R voice-allocation route.