The known-answer sales example
The source contains 14 fictional records. Data record 5 exactly repeats record 4. The prepared set has 13 lines. The week starts October 26, 2026 and ends before November 2. Eligible status is Shipped. This leaves nine lines, eight distinct orders, 26 units, RM286.60 gross, RM19.00 discounts, and RM267.60 net.
The line arithmetic is checked independently in JavaScript and in the exported workbook's calculation engine. A quantity edit and a date-window edit are exercised and restored. The workbook is rendered for layout review, and identifiers are inspected as text in the exported file. A byte-identical copy was also opened in Excel for Windows: changing Orders G6 from 2 to 3 recalculated net sales to RM280.10 and units to 27 while distinct orders remained eight. The public workbook keeps the original quantity and RM267.60 result. Download the expected results.
The browser tools
Parser checks cover quoted commas, escaped quotes, multiline records, byte-order marks, semicolons, tabs, malformed quotes, mismatched widths, and size limits. Comparison checks cover changed values, column order, repeated keys, blanks, and exact matching. Date checks cover ambiguity, fixed conventions, leap days, invalid dates, and the supported year range. Sales-report checks reconcile every source record and verify exact cent arithmetic, known duplicates, conflicting line keys, invalid amounts, zero-price lines, and date boundaries. The comparison export retains field names and every repeated source record.
The tool pages use bundled local JavaScript and have no input-upload endpoint, advertising script, or analytics integration. File text is read in browser memory. Review downloads apply an apostrophe to formula-like fields and retain unresolved findings. The original input file is not modified.
What the verification does not establish
Formula-engine checks do not prove every menu path, Power Query step, or Excel version behaves identically. Power Query instructions refer to Microsoft's documentation and require checking in your installed version. These examples do not measure real-world AI speed or accuracy. The method-comparison log is blank for your own observed measurements.
Corrections and changes
Send the URL, exact input or fictional reproduction, expected outcome, and observed outcome to [email protected]. Do not email private customer exports. A meaningful change should update the affected guide, its sample results, and the associated checks together.