What is data provenance? A plain-English guide for spreadsheet users

Someone on the Monday call asks where the 128 came from. You know it came from the invoices sheet. But which copy? Were the test accounts in or out? Did you delete the subtotal rows before you counted, or after? 20 minutes later, you're rebuilding the number to find out.

That question has a name: data provenance. Here's what it means, what a minimal record looks like for one number, and how to keep one in Excel and Google Sheets today.

What is data provenance?

Data provenance is the history of a piece of data: where it came from, what was done to it, by whom, and when. For a number on a dashboard, it's the trail from the cells in a spreadsheet to the figure on the slide.

One standard definition comes from the W3C, the body that publishes web standards. Its PROV family of documents, published in 2013, defines provenance as information about the entities, activities, and people that produced a piece of data, which you can use to judge how far to trust it. PROV's data model breaks that into 3 building blocks.

A number is never just a value. It's a thing produced by an activity that someone was responsible for. Name all 3, and you can answer the question from the Monday call.

How is data provenance different from data lineage?

The terms overlap, and standards don't draw the line in the same place. NIST's Privacy Framework uses the narrow sense: provenance is metadata about where data originated, and lineage is the history of processing of a data element, including the flows it passed through and the actions performed on it. W3C PROV uses the broad sense, where provenance covers the processing too.

In data platforms, lineage has a more specific meaning. Google Cloud's documentation, for example, describes it as a visual map of where data comes from, where it travels, and what changes along the way, tracked between data assets such as tables. It records metadata about the movement, not the data itself.

If you work in spreadsheets, this is a useful way to separate them:

If you run a data warehouse with dozens of pipelines, you need lineage. If you send a weekly report from a spreadsheet, you need provenance first, and that's what the rest of this guide covers.

Why does provenance matter for a dashboard number?

Because the same file can honestly produce different numbers. Say your invoices sheet has 1,284 rows: 3 are subtotal rows someone added for a monthly review, 9 have a blank amount, and 2 are extra copies of invoices already in the sheet. Sum the Amount column with the subtotals still in, and every invoice under them is counted twice. Leave the duplicates in, and the invoice count is 2 too high. Each version is correct for its own steps. Without a record of the steps, nobody can tell which one reached the slide.

Here are the gaps provenance closes:

• Different steps. Two people clean the same file in different ways, and both report "active clients." Why dashboard numbers don't match covers 9 common causes.
• A changed definition. "Active" meant any invoice last quarter and paid invoices this quarter, and nobody wrote down when it changed.
• Old data. The number is right for the data, but the data stops 2 weeks ago.
• No way back. Last month's number can't be reproduced because the file has changed since.

Provenance doesn't stop these from happening. It makes them visible, so the conversation moves from "is this right?" to "which version is this?"

What does a minimal provenance record look like?

For one number, you need 6 facts. Here is a record for a hypothetical number, Active clients, October: 128, built from the sheet above.

A complete record passes 2 checks. First, the counts add up: 1,270 used plus 14 left out is 1,284, the whole sheet. If they don't, some rows went missing without a reason. Second, a colleague can reproduce 128 from the same file using only the record, without asking you.

Keep the record next to the number, in the same file, not in an email or in your head. It's close to what we call a receipt. Anatomy of a receipt walks through each part and the mistake it prevents.

How do you keep provenance in Excel and Google Sheets today?

You don't need new software. These 5 habits cover most of the record.

Add a sources tab

Add a tab called Sources with one row per reported number and one column per fact in the table above. Update the row each time you update the report.

Move left-out rows instead of deleting them

Deleting rows destroys the evidence. Cut them to a Left out tab with a Reason column (subtotal, blank amount, duplicate). Then anyone can check that rows used plus rows left out equal the original total.

Put a note on the number

In Excel, notes are meant for annotations and threaded comments for discussions. Add a note from the right-click menu with New Note, or on the Review tab under Notes. In Google Sheets, Shift + F2 inserts or edits a note on Windows and Mac. One line is enough: "See Sources, row 3. Data through Oct 15."

Name the version you reported from

In Google Sheets, open version history from Last edit at the top right. Name the version you reported from, such as "Board report, Oct 16," so you can reopen that exact state. You can name up to 15 versions per spreadsheet. Google says revisions may occasionally be merged, and naming a version helps keep it from being merged. To see who changed one cell, right-click it and choose Show edit history. Google notes that some changes might not show up there, such as changes made by formulas, added or deleted rows and columns, and format changes.

In Excel, version history works for files stored in OneDrive or SharePoint in Microsoft 365, and a personal Microsoft account keeps the last 25 versions. On the Review tab, Show Changes lists who changed which cells and when, with the previous value, for the past 365 days.

Version history alone isn't provenance. It shows who changed the file and when, not which rows a number used or why rows were left out.

Let Power Query record the steps

In Excel, Power Query writes the steps down for you. Every transformation appears in the Applied Steps list in the Query Settings pane (on the View tab, select Query Settings if the pane is closed). Selecting a step shows the data as it was at that point. Right-click a step and choose Properties to give it a clear name and a description, such as "Removed 3 subtotal rows." The list becomes the Steps line of your record.

In Google Sheets, the closest habit is to clean data with formulas on a separate tab instead of editing rows by hand, so each step stays visible.

Where Every Number fits

On an Every Number dashboard, every number carries a record like this. We call it a receipt, and its 6 parts are the definition, filters, rows used, rows left out with reasons, source file, and the date the data runs through. Today it's a $99 pilot: you send one spreadsheet, and a person builds your dashboard by hand, following the same rules the product will automate after launch. See how it works or browse the sample dashboards.

It isn't the right tool for everything. If you need lineage across a data warehouse, look at a lineage or catalog tool. If your company requires a certified audit trail, know that we're a small studio with no SOC 2 or ISO 27001. And if a sources tab and named versions already answer your team's questions, keep doing that.