geometric-closed-loop-eval-edit-navigate
OUT derived (depth 8)
Created 2026-08-25T03:50:13+00:00 · Reviewed 2026-08-25T04:02:18+00:00
The MTEB evaluation coordinate system, ROME's rank-one editing, and SAE feature navigation form a single closed geometric loop: evaluation identifies the whitened directions to read, ROME modifies one whitened direction to write, and SAE neighborhood traversal navigates between whitened directions—each operation is a different linear functional on the same covariance-geometric space, unified by the Riesz map.
Justifications
SL — The three operations (read via MTEB metrics, write via ROME rank-one update, navigate via SAE neighborhoods) are individually validated but together form a closed loop only when the Riesz map provides the canonical isomorphism linking them. Retracting any one breaks the loop: without the evaluation coordinate system there is no direction to edit; without the editability proof the evaluation is disconnected from modification; without the navigation mechanism the space is static.
Antecedents (all must be IN):
- OUT evaluation-geometry-is-editing-coordinate-system — The optimal evaluation metric for an embedding model (cosine/Spearman pipeline) is simultaneously the optimal coordinate system for specifying knowledge edits, because both are readouts of the same universal covariance geometry
- OUT sae-neighborhood-as-polytope-navigation — SAE feature neighborhoods (e.g., Golden Gate Bridge → Alcatraz → San Francisco → California) are the operational navigation algorithm for the categorical polytope geometry: each SAE feature is a polytope vertex, the neighborhood structure is the polytope edge adjacency, and cross-model universality confirms this polytope is a shared semantic object rather than a model-specific artifact.
- OUT riesz-map-as-unified-operational-framework — The Riesz map under the causal inner product is the unique canonical object that simultaneously defines the evaluation coordinate system (cosine/Spearman as inner-product measurement), the editing mechanism (ROME's key-value bridge via pre-/post-activation isomorphism), and the semantic equivalence of embedding and unembedding—making all three operations different projections of a single canonical structure.