Search papers, labs, and topics across Lattice.
This paper introduces SAKIKO, a mechanistic auditing framework for evaluating internal interventions in tool-using LLMs, by formalizing representation repair through directional error discovery and router-conditioned intervention. SAKIKO reveals that while channel-keyed interventions yield net gains in most models, these gains often come with significant collateral damage to correct decisions, and promising point estimates can be misleading due to finite-sample uncertainty. The framework establishes that outcome-resolved adjudication is crucial before claiming successful internal repair, highlighting the gap between behavioral change and genuine functional improvement.
Internal interventions that boost tool-use accuracy often corrupt over half of baseline-correct decisions, revealing that behavioral improvement doesn't equate to functional repair.
Before invoking external tools, an agentic LLM must select among a K-way action space: executing a call, seeking clarification, answering directly, or declining. While internal activation steering can alter these pre-execution decisions, conventional aggregate metrics obscure where altered states land and what collateral damage they inflict. We present SAKIKO, an auditing framework that formalizes representation repair via directional error discovery, router-conditioned intervention, destination-resolved verification, and prospectively frozen statistical licensing. Across seven LLMs on When2Call and MetaTool, channel-keyed interventions induce direction-specific net gains in five models; across three sealed evaluations, none of 59 budget-matched random directions matches calibrated target gain. Crucially, destination auditing shows that behavioral movement does not equal repair: an intervention achieving +55 net gain corrupts over half of the baseline-correct decisions it touches, and promising point estimates on Qwen3-4B and Gemma-2-9B are formally declined due to finite-sample uncertainty. SAKIKO establishes the necessity of outcome-resolved adjudication before claiming internal repair. Code: https://github.com/ruizheliUOA/mechanistic-tool-use-llm.
