Recurring-timer condition accuracy

Improve exact condition accuracy for recurring-timer workflow rules, including eligibility trees and multi-phase timing conditions, where approximately 40% of evaluated cells still contain errors.

Background

The system substantially reduces syntax, omission, and structural-emission failures by translating natural-language rules into an intermediate representation and compiling that representation deterministically into workflow JSON. However, the error analysis reports persistent semantic failures in the recurring-timer family.

These failures involve incorrect eligibility trees and dropped phases rather than merely invalid serialization. Because they remain after the proposed decomposition, recurring-timer rules constitute a concrete unresolved challenge for accurate natural-language workflow generation.

References

Within the recurring-timer family, \sim$40\% of cells still miss exact conditions---real model errors (wrong eligibility trees, dropped phases) corroborated by lower judge scores, not syntax or scoring artifacts---and they define the open challenge.

— Generating Workflow DAGs from Natural Language with Non-Reasoning LLMs  (2608.30250 - Iyer et al., 31 Aug 2026) in Section 6, “Error analysis”