How to compare two CSV files and see what changed

diff old.csv new.csv and git diff compare a CSV line by line, from top to bottom. If an export comes out in a different order, almost every line looks changed, and an edited row shows up as one line removed and one added.

Matching rows on a key column such as order_id or email fixes both problems. Each record is found wherever it sits in the file, and every difference lands in one of three groups: keys that are new, keys that are gone, and keys whose values changed.

rows.page compares versions of one hosted dataset: you push the old file, then push the new file to the same link. Name the key column on the first push, and use the rpt_ token it prints for the second:

curl -T old.csv -H "X-Rows-Id-Column: order_id" https://rows.page
curl -T new.csv -H "Authorization: Bearer rpt_..." https://rows.page/<id>

1. The first push creates v1 and prints the link (https://rows.page/<id>), an owner link, a summary line and the rpt_ token.
2. The second push adds v2 to the same link. Its reply lists the change in rows and columns, with a link to the diff.
3. Open https://rows.page/<id>. The banner reads "+N new, N gone, N changed vs v1", matched by order_id.

A row counts as changed when its key is in both versions and any shared column differs. Without a key, the banner says "No unique column to match rows, so changes show as new and gone." and suggests columns to pick. You can also set the key in the address with #key=<column>.

The banner shows counts, not a row-by-row list or highlighted cells. If the previous version is over 64 MB, the comparison waits until you click "See what changed". To compare against an older version, add #diff=<n> to the address.

• No account: up to 20 versions per link, and the link expires after 7 days.
• Free ($0): sign in and choose Keep on my account to keep up to 10 datasets; the last 3 versions stay visible and older ones are purged after 90 days.
• Pro ($12 per month): 100 datasets, private datasets and 90 days of version history.
• Max ($39 per month): 1,000 datasets, private datasets and 1 year of version history.

A single push takes up to 100 MB on every plan. Pro and Max take files up to 500 MB and 2 GB through multipart uploads.

rows.page can't compare two files on your computer. To keep both files local, run these queries in the DuckDB command-line tool. Both expect the same columns in both files.

Rows that differ: EXCEPT

SELECT * FROM 'new.csv' EXCEPT SELECT * FROM 'old.csv';
SELECT * FROM 'old.csv' EXCEPT SELECT * FROM 'new.csv';

The first query lists rows that are new or edited, as they look now. The second lists rows that were deleted or edited, as they looked before. Row order doesn't matter, but an edited row appears in both results.

Changes by key: FULL JOIN

SELECT order_id,
  CASE WHEN o.order_id IS NULL THEN 'new'
       WHEN n.order_id IS NULL THEN 'gone'
       ELSE 'changed' END AS change
FROM 'old.csv' AS o
FULL JOIN 'new.csv' AS n USING (order_id)
WHERE o IS DISTINCT FROM n
ORDER BY order_id;

This lists every key that differs with its change, the same three groups as the rows.page banner. Add o.*, n.* after AS change to see the old and new values side by side.

rows.page doesn't read .xlsx or .xls files. If you push one, you get "This looks like an Excel or zip file. Export it as CSV and push again."

In Excel, save the sheet from each workbook with File > Save As as CSV UTF-8. Then push the old CSV as v1 and the new one as v2 with the curl commands above, or compare them locally with DuckDB.