manufacturing-failure-reason-codebook-normalization

The Skill normalizes failure reasons by matching raw text against product codebooks using segmentation, semantic matching, and confidence calibration. It returns a single pred_code and pred_label, or UNKNOWN when evidence is weak.

Cross-station consistency is maintained by applying station validation and evidence-based scoring during defect classification. This ensures ambiguous language across different testing stations maps to unified codebook entries.

Defect reports with typos and ambiguous language are classified through text segmentation and semantic matching against a product codebook. This fixes misinterpretations by assigning the most relevant failure code based on calibrated confidence.

When a defect report has weak evidence for any failure code, the normalization process returns UNKNOWN instead of guessing. This deterministic approach uses calibrated confidence to ensure only reliable code mappings are output for downstream analytics.

Tie-breaking conflicts during failure code assignment are resolved deterministically using evidence-based scoring and station compatibility checks. This ensures the single most relevant pred_code is selected when multiple codebook entries match the raw failure text.

Yes, failure codebook normalization supports QA trend analysis across multiple products. It standardizes testing engineers' failure reasons into unified codes, enabling consistent reporting and reliable downstream analytics across different stations.