The Lever Is Late: Causal Control of Long-Horizon Agent Termination Lives in a Task-Matched, Late Action-Commitment Block

OpenBiome

Indexed indatacite

Abstract

The first POSITIVE of the five-part WANDERING arc on long-horizon coding-agent failure (agents that keep acting but never emit the terminating 'finish' tool call). The arc established that the agent's 'task-done' verdict is linearly decodable (AUROC 0.81-0.91) yet causally inert: no residual injection rescues termination, and clamping the exact, named SAE 'done' feature moves the probability of finishing by -0.001. This paper localizes where termination control actually lives. On 99 Qwen3.6-27B SWE-bench Pro trajectories, reconstructed faithfully at the decision point and gated for behavioral fidelity (P(finish): SUCCESS 0.59 >> WANDERING 0.07 >> LOCKED 0.005), a layer-resolved logit-lens shows the finish…

Citation impact

20
total citations
FWCI
202.34
Percentile
100%
References
6
Too recent for citation history.

Authors

1

Topics & keywords

Keywords
  • Lever
  • Control theory (sociology)
  • Arc (geometry)
  • Block (permutation group theory)
  • Point (geometry)
  • Layer (electronics)
  • Control (management)
  • Feature (linguistics)
UN Sustainable Development Goals
  • Peace, Justice and strong institutions
No related works found for this paper.