Establish whether curriculum content alone determines behavior or prompts can compensate for missing cases

Determine whether curriculum content is the sole determinant of selective-withdrawal behavior in fine-tuned language agents and whether prompt instructions can compensate for missing counterexample cases.

Background

The paper interprets the survivor-preservation and composition-narrowing results as evidence that training curricula may need counterexamples requiring opposite decisions. However, the bundled curriculum changes and single-run design do not isolate curriculum content from other factors. The authors therefore leave unresolved whether curriculum coverage is solely responsible for the observed behavior or whether sufficiently explicit prompting can compensate for omitted cases.

References

The present study does not establish that curriculum content is the sole determinant of behavior or that prompt instructions can never compensate for missing cases.

— Learning When to Withdraw: Curriculum Coverage and Rule-Specific Failures in Fine-Tuned Language Agents  (2610.21045 - Abouzahra, 15 Oct 2026) in Section 6, “Curriculum coverage as a hypothesis”