SARC-DQ reveals 60% of agent AI errors come from invisible data defects

A new preprint by Gaston Besanson introduces SARC-DQ, a runtime data-quality gating mechanism for agentic AI that tackles "silent evidence defects"—metadata errors like stale prices or superseded records that look perfectly well-formed. In a priced replenishment benchmark, a competent agent silently converts an injected metadata-borne defect into a wrong action about 60% of the time, with zero data-quality flags and behavioral doubt markers at chance (AUC ≤ 0.50). The rate is flat across four model tiers spanning roughly 15x in inference price: capability does not buy skepticism.

The proposed solution, a metadata-aware pre-action gate with downstream-only remediation, recovers the loss fully on the signals its predicates cover and not at all on those they miss. A model-free oracle derived from the task's decision geometry tracks measured rates with MAE 0.015 (Pearson r = 0.876, interval coverage 15/16 cells). The paper argues that evidence integrity is a distinct systems axis from model capability, and mitigation depends on enforcement placement and predicate coverage. Code and a deterministic analysis pipeline are publicly available.