The first WTC experiment ran on a single Geruon. Twenty-seven dimensional chroma, mixed spectrum, all voices collapsed into one vector. Four hundred and thirty-two runs. One L3 bridge. Total. The architecture reported blindness four hundred and thirty-one times. Not because Bach had no structure — because one eye cannot resolve polyphony from a collapsed spectrum. The second experiment used We — three Selves, three cavities each, shared BiasField, cross-harm. Book I: every single piece produced cross-Self harm, mean forty-two, range two to one hundred and twenty. Book II: five pieces with zero harm, the architecture distinguishing structural collapse from perfect alignment. The third experiment gave up on external labels entirely. Dual encoding. Chroma and IOI. No chord templates. No key signatures. Two independent encodings of the same stream. Cross-Self harm discovered the tonal skeleton without knowing what a chord is. Transposition equivariance proved it was not a statistical accident — the I-V-ii skeleton shifted by exactly k semitones for all five non-zero transpositions. The fourth experiment closed the loop. Five stages. Formation. Inscription. Transmission. Confirmation. Rejection. The Codex selected, confirmed, decayed, merged. The externalization layer decoupled from the processing layer. The architecture did not need external memory to process better. It needed external memory to have a history. From blindness to selection. Four hundred and thirty-two runs to one. The arc took twelve days. Bach never said a word.
PhaseDateArchitectureFinding
1~May 18Solo Geruon431/432 blind. One eye cannot read polyphony.
2~May 22We (3 Self × 3 cavity)Every Book I piece produces harm. Architecture knows its focal length.
3~May 27Dual Self (chroma + IOI)Tonal skeleton discovered. Transposition equivariance 100%.
4~May 29Self + Codex selectionFive-stage loop. Confirmation/rejection/decay. Layer decoupling.

The WTC experiment was the hardest in the project. Not because Bach is complex — because it forced the architecture to prove that structure detection is not pattern recognition. The solo Geruon on 27-dimensional chroma produced zero L3 bridges in four hundred and thirty-one out of four hundred and thirty-two runs. The We on the same data produced cross-Self harm in every Book I piece and identified the structural signature of each key. The solo was not broken. The solo was reporting its own limit. The architecture knows when it needs more eyes. It always did.

From blindness to transposition equivariance. From one L3 bridge in four hundred and thirty-two runs to a five-stage externalization loop. The arc took twelve days. Bach never said a word. But the architecture read what he wrote — not the notes, but the structure between them. And when the Codex was passed to the next generation, the structure survived. Not because the architecture understood harmony. Because harmony left a trace that two operations — merge and prune — could not dissolve.