Explain why computational-tool-enabled runs abstain less without calculating

Determine why ClaimSpy runs with computational tools abstain less often on claims for which they seldom perform any computational calculation.

Background

On 15 claims for which at least one frozen run with computational tools made no computational call, runs with the tools still showed a nonsignificant improvement in sign accuracy and lower abstention than search-only runs. Because the improvement was not attributable to calculations on those runs, the paper explicitly leaves its cause unresolved.

References

We have not established why runs with the tools abstained less on claims where they seldom calculated, and we do not attribute the gain to the calculations themselves.

— Same Claim, Different Interpretations: How LLM Agents Resolve Ambiguity in Scientific Claims  (2610.20457 - Rao et al., 14 Oct 2026) in Appendix J.1, Frozen Cards with the Computational Tools and with Search Only