Dataset Diff and Reconciliation

// TWO TABLES • ROW ORDER • SORTED • EXACT VALUES

An exact comparison of the values you selected, under the import rules shown. Not an audit, not a reconciliation sign-off, and not a check of formulas or formatting. A match says the selected cells hold the same stored values; it says nothing about sheets, ranges or columns you left out, about how a value was calculated, or about whether either table is right. Use at your own risk. Open for the full scope limits.

What this tool actually does

Use at your own risk. It reads two tables, lines their columns up by exact header name, and compares the stored cell values twice: row by row in the order they arrived, and as whole rows with every column sorted. It is provided for convenience only and is not a substitute for reconciliation procedures, audit work, database tooling, or review by someone responsible for the data.

What it does not compare

  • Sheets, ranges and pairs you did not select. "All selected pairs match" is the whole claim.
  • Formatting, number formats, comments, charts, macros, connections and other workbook metadata.
  • Formulas. A formula cell is compared by the result Excel stored the last time it calculated. Nothing is recalculated, and a formula with no stored result makes the pair incomplete.
  • What a cell displays. A stored 1.234 shown as 1.23 is 1.234 here, and a date is the serial number Excel stores.

Import rules decide what "equal" means

CSV has no cell types, so a column is read as numbers only when every value in it is a plain number with no leading zero; otherwise it is text, and text 1 is not number 1. Separators, header rows, column types and how empty fields are read are settings you can see and change; the report records them. Excel writes displayed text when it saves a CSV, so a workbook compared with its own CSV export will differ wherever formatting changed the text.

Sorted differences are not record matches

The sorted check lists rows found only on one side. A record that changed shows up as one row missing on the left and one on the right; the tool does not decide that they are the same record, and it does not know which copy of a duplicated row is the extra one.

Limits and incomplete runs

Each source is limited to 40 MB, 200,000 data rows, 256 columns and 2 million selected cells. A source over a limit is refused, never compared in part. A stopped, refused or unreadable pair is reported as incomplete, and an incomplete run never shows a match.

Local processing with browser limits

Files are read in your browser and are not intentionally uploaded by this page. Browser extensions, local policies, crashes, memory pressure and device behavior remain outside this tool's control. The exported report contains values from both tables and lands unencrypted in your downloads folder.

Local is about transmission, not permission. It says nothing about whether you are allowed to open the files you just opened.

Never use this for

  • Signing off a financial close, a migration, a data load or any reconciliation someone else will rely on, without the procedures that process requires.
  • Audit evidence or a compliance record that two datasets agree.
  • Concluding that either table is correct, complete or free of duplicates.
  • Regulated data on a machine or browser you do not control, including protected health information under HIPAA, cardholder data under PCI DSS, and personal data under GDPR.

Verify independently

Before acting on a result, confirm the selected ranges, header alignment, column types and row counts against the source systems with trusted tools. Those systems control; where this page disagrees with them, this page is wrong.

1. Choose two sources

Left

Or paste a table

No source yet.

Right

Or paste a table

No source yet.

2. Pair the tables

Each pair compares one range on the left with one on the right. A blank range means the sheet's cells that hold values; the first row of a range is its header row. Sheets are paired by name only, never by tab position.

IncludeLeft sheetLeft rangeRight sheetRight rangeStatusRemove

3. Compare

How to use

The two checks

Same row order compares row 1 with row 1, row 2 with row 2, and so on, cell by cell. Sorted all columns asks whether the two tables hold the same rows the same number of times, as if both were sorted by every column first. If the first check fails and the second passes, only the row order differs. One inserted row shifts every row after it in the first check; read the second check before concluding that many cells changed.

Columns are matched by header

A column on the right is found by its exact header text, so reordered columns still line up. Qty, qty and Qty (with a space) are three different headers; the results point out names that differ only in case or spaces. A header on one side only stops the comparison for that pair, rather than quietly dropping the column.

Duplicates count

If a row appears three times on the left and twice on the right, one copy is unmatched. The report lists every source row holding that pattern, because which copy is the extra one cannot be known.

Excel stores, it does not display

Values are compared as Excel stored them. Two cells both showing 1.23 can differ (1.234 and 1.233), and two cells showing a date differently can match. A date is a serial number; workbooks using the 1904 date system store different serials for the same calendar date, and the report says so. Hidden rows and columns inside a range are compared.

Dates in CSV files

A CSV date is text until you say otherwise. Set a column to Date, pick the format its values use, and pick the date system of the workbook you compare it with; each date then becomes the serial number Excel stores for it, time of day included. If any value in the column does not fit the format exactly, the whole column stays text and the results name the values that did not fit. Nothing is guessed, and dates are never read automatically.

File formats

XLSX, XLSM, XLSB and XLS read the same stored values; macros in an XLSM file are never run, and password-protected workbooks are refused because this page never asks for a password. ODS loses some detail on the way in: hidden sheets, rows and columns are not detected, a stored empty string reads as blank, dates before 1 March 1900 read one day later than Excel's serials, and a file holding error cells such as #N/A cannot be read. A pair that uses ODS carries those notes in its results.

Row numbers

References in the report are to the source: a worksheet cell such as C8, or a CSV record and field. A record that spans several lines because of a quoted line break is still one record. The order rows appear in the sorted section is only a display order.

Exports

The .txt export is the full report. The two CSV files list the same differences one per row for a spreadsheet: every changed cell with both values and references, and every row found on one side only with its copies and source rows. Text that a spreadsheet would run as a formula (starting with = + - or @) is written with a leading apostrophe, and that row is marked. The .json file keeps every value exactly, with its type. Print shows the summary and the preview, with the shown and total character counts; it is not the complete report.

Saving settings

Save settings writes the pairs, ranges, text encodings and import settings to a small JSON file. It holds no file names and no values from your tables. Load settings applies the file to the sources already loaded, and to the next file you load on each side; then choose the files again.

Example files

The examples above load small files that each show one thing. To open a pair in Excel, download display-left.xlsx and display-right.xlsx: both show 1.23 in cell B2, and the comparison says why they still differ.

About this tool

Workbooks are read by SheetJS Community Edition 0.20.3, served from this site. Comparison, CSV parsing and the report are written for this tool.