Can geometry predict the working set?
Observed together. Useful together?
Does residual-stream geometry predict which parts of a model will be active or resident together? The existing research note records co-activation as observed and useful locality prediction as untested.
Can residual geometry predict physical working-set behaviour?
CURRENT POSITION
The existing working note says co-activation has been observed, but has not been shown to predict anything useful. It records observing expert co-activation across related prompts on Kimi K3; locality prediction remains untested.
PREDICTIVE LOCALITY IS NOT ESTABLISHED
Recorded observation: expert co-activation
Open question: useful physical working-set prediction
Present is not the same as predictive.
predictive locality is not established — recorded observation: expert co-activation; open question: useful physical working-set prediction. Present is not the same as predictive.
SOURCES & PROVENANCE
- Existing residual geometry question
residual-geometry-question.ts, recorded 26 August 2026. This source is a working note, not a published experiment.
AUTHOR / CHRIS HAY · VERSION / 0.1
REFERENCE THIS DRAFT
An unpublished working record. These references identify the draft and omit a publication date. They become version-specific publication citations when the record is released.
Chris Hay. Can geometry predict the working set? [Unpublished draft, version 0.1]. https://chrishayuk.com/research/residual-geometry