Similarity-based locality of relevant helper assertions

Ascertain whether helper assertions relevant to repairing a given Dafny lemma are more likely to be found within lemmas that are structurally similar to the target lemma in the same codebase, thereby motivating selection of in-context examples based on lemma similarity.

Background

Selecting useful in-context examples is critical because codebases may contain hundreds of assertions, and providing too many examples to the LLM is infeasible. The paper introduces a lemma similarity metric to identify structurally similar contexts and hypothesizes that assertions from similar lemmas will better guide synthesis.

This conjecture underpins the example selection strategy in Laurel: by choosing assertions from lemmas similar to the target lemma, the tool aims to improve the likelihood that generated assertions match the structure and needs of the failing verification condition.