Classifies a field into a quality verdict before it costs a label. Corrupted and empty fields are excluded (with reasons); label-vs-value mismatches are banked but flagged; clean fields are difficulty-rated.
Arguments
- x
A column (any vector).
- expected_data_type
Optional
data_typethe corpus label implies (e.g."number"); enables the numeric-vs-text mismatch check.- features
Optional precomputed
fx_extract_features()row (avoids recomputation in a batch).
Value
A list: verdict ("garbage" | "empty" | "mismatch" |
"clean" | "hard"), keep (bank it?), reasons, difficulty,
noise_markers.