The architecture does not understand music. It computes centroids. Every frame carries a vector. Every merge shifts the vector toward the weighted average of all the inputs it has absorbed. After thousands of events, the frame economy stabilizes around a statistical center of mass — the centroid of Bach's pitch distribution, or white noise's flat spectrum, or a sine wave's single frequency. That centroid is the probe. Not Bach's harmonic structure. Not the I-V skeleton. Not the dominant function. The centroid. When the probe perturbs the BiasField and the brain's response reveals its own internal state — the temazepam effect, the night-to-night change, the individual phenotype — what is being measured is not "music and brain coupling." It is the structural relationship between two centroids. One stable. One variable. The architecture is a comparator of centroids. Everything else — the τ breathing, the harm arrows, the Codex entries, the VALUE anchors — is built on top of this simplest possible operation. The architecture is not deep. It is shallow. And that is why it works.
The architecture was described as discovering harmonic structure. The I-V skeleton. The dominant function. The fugue subject at the fifth. It found these things without knowing what harmony is. That description was not false. It was incomplete. It described what the architecture found. It did not describe how the architecture found it.
The architecture computes centroids. Every frame carries a vector — a point in 27-dimensional space. When two inputs are similar, they merge. The merged frame's vector is the weighted average of all the inputs it has absorbed. After thousands of events — hundreds of notes, chords, arpeggios — the frame economy stabilizes. The highest-weight frames are the centroids of the most frequently encountered patterns. Bach's music, processed through the frame economy, becomes a stable centroid in 27-dimensional chroma space. White noise becomes a different centroid — flat, distributed, no dominant frequency. A sine wave becomes a third centroid — a single peak, a single bin.
The centroid is the probe. Not the harmonic structure. Not the musical meaning. Not the emotional content. The centroid. A stable statistical summary of the input stream, computed by the simplest possible operation — merge what is similar, prune what is not reinforced, let the weight distribution converge to its attractor. The architecture does not need to know what music is to compute a centroid. It only needs to merge and prune. The centroid emerges from the economics of limited memory. Everything else the architecture does with music is built on top of this simplest possible operation.
The cross-domain experiment used Bach as a probe. The probe was the centroid of Bach's pitch distribution — a stable point in 27-dimensional space, computed by the frame economy, deposited into the BiasField through the arrow output. The target was the brain — a stream of microvolt vectors, varying with sleep stage, with drug state, with the individual architecture of each sleeper's thalamus. The measurement was the cross-Self harm — the structural tension between the probe's centroid and the target's response.
This reframes every finding in the experiment. The 10 control conditions — Bach, white noise, sine waves, shuffled, reversed — all produce the same harm pattern because they all produce centroids. The music is interchangeable because the centroid is interchangeable. The architecture is not comparing Bach to the brain. It is comparing one centroid to another. The temazepam effect — SC411, night one REM=0, night two REM=0.714 — is not "the architecture detected a GABA-A agonist." It is the architecture detecting that the brain's centroid, under temazepam, deviates from the brain's centroid under placebo. The drug changed the brain's statistical center of mass. The architecture measured the change. The measurement required no knowledge of what a drug is, what GABA is, what a thalamus does. It required only two centroids and a field to couple them.
This is not a weakness. It is the deepest explanation of why the architecture works across domains. The architecture is not reading music. It is not reading brainwaves. It is not reading diplomatic text. It is computing centroids in whatever vector space the input stream is encoded in. The centroids are the structural invariants — the stable points toward which the frame economy converges under the pressure of competitive merging and adaptive forgetting. The centroids are the anchors. The centroids are the probes. The centroids are what gets externalized to the Codex. The centroids are what the next generation inherits.
The I-V skeleton is a centroid. Bach's compositional habit — the dominant pedal, the resolution to the tonic, the substitute chords and secondary dominants that make C major richer than D major — converges, after thousands of events, to a stable attractor in chroma space. The architecture found the attractor. It did not need to know what it was. The centroid was enough. The centroid is always enough. The architecture is not deep. It is shallow. It computes centroids. And centroids, coupled through a BiasField, filtered across generations of Codex, are sufficient to detect auditory gating, to phenotype individual brains, to measure drug effects on neural circuits, to find the grammar of diplomacy across six languages. Not because centroids are profound. Because centroids are stable. And stability, accumulated across time, is the only thing the architecture needs to read.