rcode-ocr-consistency

What problem does it solve?

This Skill helps teams detect and prevent OCR pipeline drift so text extraction does not silently change across runs, model updates, or infrastructure changes.

Core Features & Use Cases

• Language routing checks: Verifies that document language is classified before OCR and that low-confidence cases are surfaced for review.
• Model determinism checks: Confirms OCR model versions, Triton configuration, and decoding settings are pinned to avoid silent output changes.
• Ground-truth validation: Measures accuracy against verified documents and blocks release when regressions exceed the defined threshold.
• Observability and replay: Ensures traceable logging, alerting, and deterministic reprocessing from archived inputs for incident investigation.
• Use case: A team notices bilingual PDFs returning different text after an infrastructure rollout and uses this Skill to isolate whether the regression came from routing, model drift, or deployment configuration.

Quick Start

Ask the skill to audit your OCR pipeline against the six consistency checks and report the findings, regressions, and required memory bank updates.