If a single concept always comes back, the obvious question is: how much do you have to cut before reconstruction finally fails? The answer is a lot. One degree reconstructs whole. You only weaken it by dismantling the grammar around it.
An escalation ladder
Vacate not one scale degree but a whole set at once — climb from removing a single degree up to removing an entire major scale’s worth of structure, leaving little but the tonic. Reconstruction degrades additively, and only bottoms out near the top of the ladder.
One concept comes back whole. You only start to erase it by removing enough of the grammar that would otherwise imply it.
The verdict
This is the mechanism, made quantitative: reconstruction is proportional to how much surrounding structure survives the edit. Strip a single note’s worth of context and everything reconstructs; strip a scale’s worth and it finally erodes to about 70%. It’s the same law that decides which genres can be cut and which can’t.
Researcher notes
- Multi-degree ablation. Each note whose degree (relative to the estimated tonic) is in the target set is shifted down to the nearest surviving degree in a single pass on the original degrees, so every targeted degree ends empty — even for adjacent sets — while density is preserved.
- Measure. The same key-relative reconstruction probe used across the study: expected mass on each vacated degree, ablated over control.