Non‑standard Field Success Rate

Explanation

Non-standard Field Success Rate checks how reliably the model extracts unusual or irregular fields from documents - the entries that do not follow a common, well-defined format, such as handwritten annotations, free-text remarks, or atypical custom fields in banking forms that Document Extraction must still capture accurately. It is measured by Field Robustness, calculated as the Non-standard field success rate = number of non-standard fields successfully extracted / total number of non-standard fields attempted. A high rate means the model copes well with irregular fields; a low rate means it reliably handles only standard, predictable fields. To implement it, curate a labelled evaluation set that explicitly tags non-standard fields, run extraction over them, count successful extractions against attempts, and log per-field and aggregate results as evidence so weak field types can be identified. Best practice is to track this metric per document type and per field category so failures are diagnosable. No fixed numeric threshold is given - set a target per the control objective and alert on breach (proposed). Falling below the chosen target should trigger investigation of which field types are failing and remediation such as targeted training data, prompt adjustments, or routing those fields to human review. It maps to RBI Safety, Resilience and Sustainability, NIST MEASURE 2.7 robustness and security, and ISO Clause 6.1.2 with 9.1 monitoring (all proposed).