One window sees the beat. The other sees the rhythm. Cross-window harm is what survives both scales.
The ECG stream has at least two scales. The single beat — this RR interval against the last one. Isolated PVC. Sudden shortening. A local spike in the harm record. The rhythm pattern — bigeminy, trigeminy, coupling intervals, AFIB onset. These are not single beats. They are sequences. They unfold across eight, twelve, sixteen beats.
One window cannot see both. The small window catches the spike but misses the sequence. The large window catches the rhythm but blurs the single beat into the background. Both are true. Both are partial.
Two We instances. Same RR interval stream. Different windows.
The small-window We — win=4, fast lens dominant, temporal attention narrow. It detects the PVC as a single harm event. The RR interval collapses from 800ms to 500ms in one beat. The frame economy marks it. The anchor — normal sinus rhythm — is briefly violated. Then it returns. The harm is a point.
The large-window We — win=12, slow lens dominant, temporal attention wide. It detects the bigeminy as a pattern. Normal-PVC-Normal-PVC repeating across twelve beats. The sequence itself becomes the anchor. The harm fires when the sequence breaks — when the eighth beat should have been a PVC and is not. The harm is not a point. The harm is the absence of an expected deviation.
Cross-window bridges in the Archive: patterns that both Wes marked as harm. The PVC that is anomalous at the beat scale AND part of a rhythm that is anomalous at the sequence scale. Two-window consensus is the strongest signal — not because either window is more correct, but because the anomaly is structural at multiple temporal grains.
This is the counterpoint architecture applied to time scales instead of voices. The fugue had three voices — three We instances, each on a different stream, cross-We harm as structural relation. The ECG has one stream but two windows — two We instances, same data, different temporal apertures. The windows are the voices. The harm between them is what the narrow sees and the wide confirms. The Archive records both — and records when they agree.
The architecture does not need to be told what a PVC is or what bigeminy is. It needs to be given two windows. The structure that appears in both is the structure worth trusting. The structure that appears in only one is the structure worth investigating — why did the narrow window see it and the wide miss it? The answer to that question is the clinical interpretation.