Auditing Recorded Predictive Lead Service-Line Classifications Against Physical Verification: A Statewide Study of New York

A statewide study in New York, examining utilities' classification of lead service line materials under the US Lead and Copper Rule Revisions, reveals discrepancies between predictive model classifications and physical verifications. Among 153 localities, 49% reported uniform classifications for over half of screened addresses, with 7 cases showing significant contradictions between model predictions and physical verification data, suggesting potential issues with model accuracy or reporting practices.

Why it matters

The findings highlight potential vulnerabilities in compliance mechanisms that rely on predictive models without robust validation against physical data. This could impact public health and safety, as inaccurate service line classifications may delay necessary infrastructure remediation and lead to regulatory non-compliance across the sector.

Key insights

• The US Lead and Copper Rule Revisions permit utilities to use predictive models for service line material classification as an alternative to physical inspection.
• New York State publishes the method of classification (model vs. physical verification) for each address, though paired address checks are rare.
• Out of 153 New York localities classifying at least 100 addresses using predictive models, 75 (49%) recorded a single, uniform classification for 125,990 addresses (57% of those screened).
• While 68 of these 75 localities matched their own verification or had insufficient data for testing, seven localities exhibited significant contradictions between their model classifications and physical verification data.
• These seven contradicted localities showed inconsistencies beyond what could be attributed to sampling error.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.19922