Check Decimal and Thousands Separators When Importing PCB Measurements

The text “1,250” can represent different numeric values under different formatting conventions. A test file exported on one workstation may therefore be interpreted differently on another. Define numeric formatting at the interface and verify the parsed value before applying product limits.

The pandas CSV import documentation exposes decimal and thousands-separator settings separately. This illustrates why choosing a column delimiter is not enough: the text inside a numeric field also needs a defined interpretation.

Identify the source convention

Record the exporter, locale configuration, decimal separator, grouping separator if used, and scientific-notation format. Keep the measurement unit in a separate controlled field or another unambiguous part of the schema. Do not infer the convention from a single value that contains no punctuation.

Check whether the file’s delimiter conflicts with the numeric format. Quoting may preserve a decimal comma inside a comma-delimited field, but only if the parser and exporter agree on quoting. Resolve record structure before numeric conversion.

Use values that reveal ambiguity

Compare parsed values with known source values, not only with whether conversion succeeded. A parser can produce a valid number that is off by a factor of a thousand. Check plausible ranges as a secondary guard, while recognizing that some wrong interpretations can still fall inside a broad range.

Keep conversion accountable

1. Retain the original field text.
2. Apply the declared numeric convention.
3. Record conversion failures explicitly.
4. Verify units and scale before limit comparison.
5. Store the parser configuration with the imported dataset.

Do not remove all commas or periods as a universal cleanup rule. That operation can fix one convention while corrupting another. If mixed conventions occur in one dataset, identify their provenance and resolve them through explicit rules or review rather than guessing row by row.

Test the complete path after moving an exporter to another computer or changing spreadsheet settings. The same application code can encounter different presentation defaults. Include the received text, parsed number, and expected value in a small regression dataset that covers each supported exporter and numeric convention.

The goal is a stable numeric contract between the station and analysis system. Formatting should not determine whether a board passes, and an ambiguous field should remain an unresolved data issue until its intended value is established.