Residual Paving: Diagnosing the Routing Bottleneck in Selective Refusal Editing

TL;DR

This paper introduces Residual Paving, a routing-based residual editing method for frozen transformers, significantly reducing edit refusal rates while preserving benign and harmful distributions, and diagnosing bottlenecks in selective refusal editing.

Contribution

It proposes a novel routed residual editing approach that separates route selectivity from residual capacity, enabling effective diagnosis and improvement of selective refusal editing in transformers.

Findings

Reduces edit refusal from 88.6% to 4.0% on Gemma-3-4B-IT.

Achieves 95.5% benign and 87.3% harmful distribution preservation.

Oracle routing improves diagnostic scores across six backbones.

Abstract

We study selective refusal editing as a three-way control problem: induce non-refusal on designated edit prompts while preserving benign behavior and harmful refusals outside the edit set. We introduce Residual Paving, a routed residual editing method for frozen instruction-tuned transformers that separates route selectivity, whether to intervene, from residual-edit capacity, what edit to apply. An early-layer router predicts a scalar gate and expert mixture; when active, prompt-conditioned bottleneck residual experts apply later-layer residual updates while leaving the backbone unchanged. This decomposition supports an oracle-routing diagnostic where only the learned scalar gate is replaced with the held-out edit/keep label, leaving the residual editor and frozen backbone fixed. On the primary Gemma-3-4B-IT held-out split, learned Residual Paving reduces edit refusal from 88.6% to 4.0%, with 95.5% benign distribution preservation and 87.3% harmful distribution preservation. Same-protocol one-direction steering controls are much weaker on edit success, leaving edit refusal at 86.8% for Edit-target ActAdd and 78.9% for DIM-style refusal steering. The remaining failure is off-target harmful-keep degradation: harmful refusal remains below the frozen-base rate, 65.3% vs. 81.6%. Across six backbones, oracle routing improves the keep-side diagnostic score on every reported row, with median gain +12.9 pp, supporting the interpretation that learned route selectivity is the main observed bottleneck. Trajectory diagnostics on two backbones further suggest directed movement toward edit-target continuations rather than generic refusal suppression.