music/pass2/measurements/v2_sustain_and_harmony_variants/README.md
This directory holds the caci cue AFTER the sustain repair and the harmony and pokok variants, and
BEFORE the thin points. It is kept because it is a refutation rather than a step.
The hypothesis it tested: a section boundary the audio segmenter can find needs the HARMONY to move,
not only the instrumentation. Four harmony variants and four pokok variants were added, no two
consecutive sections sharing a pair.
The result: five measured sections against the first render's eight, and structural yield 0.118
against 0.226. Worse on the axis it was meant to repair.
The exploration cue received the same two changes over the same interval and improved decisively —
drops 35 to 20, structural yield 0.714 to 1.100. So the variants are not harmful in general. They
are simply not what a dense cue's segmenter is reading: in a thirty-nine-part mix the layers
carrying the harmony sit twenty decibels under the ones carrying the timbre, and a chroma-and-MFCC
self-similarity matrix reads the ensemble as one object however much its inner harmony moves.
What the refutation pointed at instead is DENSITY SWING, which is the third render's change.
Read this directory beside v1_first_render/ and the live measurements in the parent directory.
The verdict is build/audio/pass2/PROTOTYPE_VERDICT.md.