Established onExcel 365 with vendor cloud storage: change list, 60 daysGoogle Sheets, web: named versions and per-cell historyExcel 2016, local files: nothing
Both major products do keep a version history now, and the gap between what it records and what a version history in software records is the subject of this entry.
What exists
A workbook stored in either vendor's cloud service keeps previous versions and lets you restore one. Google Sheets names versions, shows a list by time and author, and will report the edit history of an individual cell. Excel offers version history for files on the vendor's storage, and since 2021 a change list for the last sixty days.
That is more than existed a decade ago and it is worth turning on. It is also a list of moments rather than a record of decisions.
What is missing, and why each absence matters
No description of the change. A version is identified by a time and a person. Why the change was made is not recorded anywhere, so a figure that moved six months ago cannot be explained without finding somebody who remembers.
No readable comparison. There is no view showing which formulas differ between two versions. Restoring means replacing the file wholesale, and finding what changed means looking.
No approval before the change takes effect. An edit is live the moment it is typed. Nothing exists between making a change and the change being in the result everybody uses.
No tests. There is no artefact that says what the sheet is supposed to produce and complains when it does not.
Why the file format does not help
A modern workbook is a compressed archive of structured text, so it can be stored in a version control system, and the comparison it produces is unreadable: a single edited cell rewrites large parts of several internal files.
Tools exist that extract the formulas to plain text so that a comparison becomes meaningful, and they are worth knowing about for anything financially significant. None is standard, and none of them is in the workflow of the vast majority of people who use spreadsheets.
What can be reconstructed by hand
A test, as a row of cells. Put the answers you already know beside formulas that compute them, and a comparison that says whether they agree. Where the sheet is right, the row reads yes across; where it is not, it does not. This is the whole of testing, adapted, and it costs five minutes.
A description, as a cell. A dated line at the top of the sheet saying what changed and why converts the version list from times into decisions. It is only as good as the discipline behind it and it is better than nothing, which is the alternative.
A boundary, as a protected range. Locking the calculated cells so that editing them requires deliberately unlocking them is not review, and it does put a step between an accidental keystroke and a changed model.
The one place this is genuinely solved
Where the calculation matters enough, it leaves the spreadsheet. A figure produced by code in a version-controlled repository has all four of the missing properties without anybody reconstructing them.
That is not a recommendation to rewrite everything, and the final entry of this site is about when it becomes the right answer. It is an observation about what is being given up in exchange for the speed of putting a number in a cell.
What we cannot verify
The version history features and their limits are documented by both vendors and can be checked directly; the sixty-day window and the storage requirement for the change list are stated by the vendor and differ by plan. The behaviour of comparison tools was established by testing on current builds and is not a claim about older versions.
What a shared file does to the question
Where several people edit the same workbook at once, the version history stops being a record of one person's decisions and becomes an interleaving of several. Restoring a version to undo one person's change undoes everybody's changes since.
Which is why the practical unit of recovery in a shared sheet is a single cell's history rather than the file's, and why the per-cell view in one of the two products is more useful in daily work than the version list beside it.
The habit that costs nothing
Name a version at every point where somebody acted on the numbers. A version named for the meeting it was presented at is findable; a version identified by a timestamp is not, six months later, by anybody.
In short
- Both products keep version history, and it records moments rather than decisions.
- There is no description of a change, no readable comparison and no approval step.
- Nothing states what the sheet should produce, so nothing can complain that it does not.
- The file is an archive of structured text, and comparing two copies is unreadable.
- A test can be reconstructed as a row of known answers and comparisons.
- Where the calculation matters enough, it leaves the sheet and gets all four properties free.