SALVE + Matryoshka SAE + the broader 2026 methodology family = H2 2026 mech-interp pluralization continues across multiple methodology axes simultaneously

SALVE provides SAE-mediated mechanistic control methodology. Matryoshka SAE provides multi-resolution feature learning architecture. Combined with the broader 2026 methodology family (Binary Sparse Coding + PRISM + multi-layer SAE + SAE-LoRA), H2 2026 mech-interp continues pluralizing across multiple methodology axes simultaneously.

SALVE SAE-latent-vector-editing methodology + Matryoshka SAE multi-resolution architecture together demonstrate H2 2026 mech-interp methodology pluralization continuing.

The methodology-axis pluralization

H2 2026 SAE methodology family now spans: continuous SAE (mainstream) + multi-layer SAE (cross-layer feature tracing) + PRISM (polysemanticity capture) + Binary Sparse Coding (discrete representations) + Matryoshka (multi-resolution) + SAE-LoRA (parameter-efficient steering) + SALVE (mechanistic control) + concept-annotation evaluation methodology. Each addresses different limitations or application axes.

The discovery-vs-steering debate continues

The discovery-not-steering position paper argued SAE methodology should reposition toward discovery. SALVE's steering-application methodology provides concrete counter-evidence that steering remains productive direction. The H2 2026 SAE methodology direction operates on both axes; the debate continues.

The procurement implication

Safety-engineering procurement of interpretability tooling should now match methodology choice to specific application requirements rather than seek single-best methodology. The H2 2026 to 2027 SAE methodology landscape supports multi-methodology procurement matching application-shape diversity.

arXiv — SALVE: Sparse Autoencoder-Latent Vector Editing (2512.15938) → · arXiv — Learning Multi-Level Features with Matryoshka Sparse Autoencoders (2503.17547) →