manufacturing-failure-reason-codebook-normalization

This skill acts as a specialized data normalization engine designed for manufacturing quality control environments. It processes raw defect logs written by testing engineers, converting unstructured, error-prone text into structured, valid codebook entries. The system is particularly effective at handling common industry pain points such as ambiguous descriptions, inconsistent terminology, improper mixing of Chinese and English characters, and cross-project writing habits. It ensures that every reported failure aligns with the pre-defined product codebook through a rigorous semantic matching and station-validation pipeline.

• Performs automated segmentation of raw failure logs into discrete, processable segments.

• Executes station-level validation, automatically rejecting codes that are incompatible with the specific test station or assembly line segment.

• Uses fuzzy matching and semantic analysis to link descriptive inputs to standard codebook labels.

• Applies deterministic tie-break logic for near-duplicate candidate codes, ensuring reproducibility in ambiguous cases.

• Generates confidence scores for every prediction, allowing the system to flag low-confidence results as UNKNOWN for manual review.

• Calibrates output confidence to ensure a clear distribution-level separation between reliable auto-matches and manual-intervention-required alerts.

• Ideal for quality assurance teams, production engineers, and data analysts managing manufacturing datasets.

• Expected inputs include raw_reason_text strings, product-specific codebook datasets, station identifiers, and associated test metadata.

• Outputs provide a standardized pred_code, pred_label, and a normalized confidence score.

• Constraints: The skill requires access to a valid product codebook and must adhere to the provided station_scope_map to maintain data integrity.

• Tips: If an entry is identified as UNKNOWN, it suggests that the provided reasoning contains insufficient cues or references to match against the current version of the codebook; in such cases, consider reviewing the source log for missing station context or non-standard abbreviation usage.