Distinguish whole-column format inference failure under strict=False
One invalid row differs from a column with no inferable format. Keep an explicit format beside actual completion status.
Inference precedes per-row failure
Separate per-row parse failures from failure to infer any format for the column. Test completion and missing representation under an approved explicit format. Teams interpreting strict=False at different stages can otherwise record an interrupted run as a successful empty result. Connect Truthound to a missing-flag comparison table prepared after parsing has completed. [1, 2, 3]
The reporter narrows the all-invalid case
Polars #29420 reports an error rather than NULL when no format is supplied and the column contains only bad date. The reporter later corrects an initial misunderstanding: if at least one date supplies an inferable format, other bad rows become NULL. Do not repeat a claim that any invalid date fails the whole column. [1]
Do not relabel failure as successful NULL output
Define allowed formats and record completion, original text, and missing flags. Do not portray an inference exception as normal NULL output; test an explicit-format path. Preserve original text and format, checking completion before inspecting row-level flags. [1, 3]
An independent missing-value expectation
The illustrative explicit format is %Y-%m-%d, which bad date does not match. After successful nonstrict conversion, required missing flag true is the independently defined state. Actual true passes; false fails. If the run raises an exception, do not manufacture a successful flag for comparison. [1, 3]
Compare flags after completion
ColumnPairValidator in Truthound Python SDK 3.3.2 compares actual_null_flag_after_success and required_null_flag prepared by the operator after completed parsing. Operators run the parser and classify errors. Core does not catch parsing exceptions, infer all formats, or guarantee date validity. [1, 2, 3]
Classify allowed formats explicitly
Use an explicit format or classify allowed formats before parsing them separately, retaining failed source text. A regression already covering the all-invalid case can reduce another audit’s value. Keep execution errors distinct from a legitimate column of NULL values. [1, 3]
Application conditions
Check parsing completion first, retaining original text, explicit format, and row-level missing flags.
Alternatives
Test an explicit format or prior classification by allowed formats, retaining failed source text.
Limitations
Flag comparison on completed output does not handle parser exceptions or infer every format.
Evidence
- Event checked
- Subject checked
- Modified
