Performance of the baseline with larger rule corpora and retrieval

Determine how the end-to-end baseline for LLM-assisted medical billing verification would perform with a substantially larger rule corpus and whether retrieval would be necessary.

Background

The paper evaluates its baseline against the decision-basis contract using a compact synthetic rule corpus containing 28 entries across two catalog releases. Because the baseline receives the catalog and performs rule checking, documentation assessment, and outcome determination in a single request, its behavior may change substantially when the rule corpus becomes much larger. The authors explicitly identify as unresolved both the baseline’s performance at that scale and the possible need for retrieval mechanisms to provide the relevant rules.

References

It remains an empirical question how the baseline would perform with a much larger corpus, or whether retrieval would be necessary.

— A decision-basis contract for auditable LLM-assisted medical billing verification: deterministic rules, verbatim evidence, and fail-closed abstention  (2609.12156 - Hölter et al., 10 Sep 2026) in Section 4, Discussion