The .xlsx file.
Also .xlsm, .xls, which are read the same way.
- Really a
- ZIP archive of XML parts
- Rows per sheet
- 1,048,576
- Columns per sheet
- 16,384, ending at XFD
- Number precision
- IEEE 754 double, displayed to 15 significant digits
- Dates stored as
- A serial number counting from 1900, or 1904 on some Mac files
- Read here as
- Direct. Every sheet, including the ones that are notes.
- Status
- Ready
- Fidelity
- Read directly
- Largest file
- 25 MB free, 250 MB once a project is bought
What it actually is
Every sheet read, including the ones that are notes rather than data.
An .xlsx is a ZIP archive. Rename one to .zip and open it and you get a folder of XML: one file per worksheet, one shared table of every string in the workbook, and separate parts for styles, charts and defined names. This is why a workbook can be read without Excel installed, and why a corrupt file is often recoverable by opening the parts directly.
It replaced the older binary .xls in 2007. The important consequence for analysis is that everything in the file is addressable: a formula, a cell's number format, a merged range and a pivot cache are all separate, inspectable structures rather than opaque bytes.
What it is not is a table. A worksheet is a grid of cells with no concept of where a dataset begins or ends, which is why a title block, a stranded subtotal and three tables side by side are all perfectly legal and all completely invisible to anything that assumes row 1 is a header.
Upload one
The analysis runs before there is anything to pay for.
You see what it found, and the evidence behind it, first.
No account needed to start. You only pay when you like what you see.
What goes wrong, and what is done about it
4 named failures, each with the repair.
The 1900 leap year that never happened
Excel's date serial numbers treat 1900 as a leap year, because Lotus 1-2-3 did and compatibility won. Every date before 1 March 1900 is off by one day. Separately, workbooks saved on older Macs count from 1904 instead, so the same serial number is four years and a day apart depending on the file.
What is done: Dates are read with the workbook's own epoch setting rather than a global assumption, and a file mixing both is reported instead of silently blended.
Numbers stored as text
A value that arrived as "1,240.00" with a trailing non-breaking space is a string. It looks identical in the cell and is skipped by every sum. The total is not wrong so much as quietly incomplete, and the shortfall is usually small enough to look plausible.
What is done: Numeric-looking text is detected per column and re-read as numbers. The columns re-read that way are named in the methodology, so you can see which totals depended on the conversion.
Subtotal rows that look like data rows
Someone adds a total under each region. Someone else later sums the whole column, and every value is counted twice. The signature is a figure that is close to double, or that exceeds the sum of the parts by a suspiciously round proportion.
What is done: One totals row at the foot of a table is dropped, and only when every numeric column in it reconciles against the rows above. Subtotals scattered through the middle of the data are not removed: they are reported as a risk to the figures, so the double counting is visible rather than silent.
15 significant digits, silently
Excel stores numbers as doubles and truncates display at 15 significant digits. A 16-digit account number, order id or barcode typed into a general-formatted cell loses its last digit permanently, and the file gives no indication that it happened.
What is done: Columns of long numeric identifiers are read as text where the values show the trailing-zero signature of this truncation, and flagged so the source can be re-exported as text.
Questions people ask
About this format, not about the product.
Do I need to tidy the workbook first?
No, and tidying it by hand usually destroys evidence. Merged cells, several tables on one sheet, a title block above the data and dates in four formats are the normal case. Finding where the real tables start and end is the first thing that happens, and every repair is listed with a before and after.
Are macros run?
Never. A macro-enabled .xlsm is read for its values; the VBA project is not executed and never will be. There is no configuration that turns that on.
What happens to my formulas?
The computed values are read; the formulas are not. What arrives is what the workbook last calculated, so a hardcoded number sitting inside a range of formulas, or a range that stops one row short of the data, arrives as an ordinary number with nothing to mark it. Those are two of the most common ways a workbook is wrong, and finding them is still work for a person.
What is the largest workbook I can upload?
25 MB without a purchase, and 250 MB once a project is bought. The larger of those is many times the size of the largest workbook Microsoft ships as a sample of a large one: its Retail Analysis model holds 1.3 million rows and does not come close. Above it, export the sheet as CSV or Parquet: both are a fraction of the size for the same rows.
Keep reading
The jobs people do with this file, and the formats beside it.