Many of the free report analyzers on marisstella.app read a CSV export: a delivery-app payout report, an Etsy statement, a Shopify payout file, a KDP royalty report. When a row can't be read (a date that isn't a date, an amount that isn't a number) the page says so: "line 3 skipped, date not readable". The idea is that you open the file, go to line 3 and see what went wrong.
Until today that line number was often wrong.
How it went wrong
The analyzers share one small CSV parser, copied into 49 files. It does two sensible things: it drops blank lines, and it keeps a line break inside a quoted field ("Joe's order\nno onions") as part of one value. Both mean a row's position in the parsed table is not its line in the file. Forty-one row loops in 34 analyzers still reported "row index + 1" as the line.
So an export with one blank line after the header, which many spreadsheet tools write, pushed every later message up by one. A note field with a line break in it did the same. On the restaurant delivery commission analyzer, this file:
Order ID,Platform,Order date,... line 1
A,DoorDash,2026-06-01 12:00,... line 2
line 3 (blank)
"B line 4
note",DoorDash,2026-06-01 12:30,... line 5
C,DoorDash,not a date,... line 6
reported the bad date on line 4, which is the start of a perfectly good order. It now says line 6.
The totals were never affected: the rows themselves were read correctly. Only the pointer back into your file was off.
The fix, and the check that keeps it
Each copy of the parser now counts physical lines as it reads, including the breaks inside quotes, and gives every row the line it starts on. The value is attached so it doesn't show up when a row is compared or printed, so nothing else changed. Each of the 41 loops reads that line and falls back to the old number only when a table didn't come from the parser.
One test fixed one copy. The other 48 needed something that runs on every pull request without anyone remembering it. A new check imports every copy of the parser, feeds it a file with a blank line, a quoted line break, a Windows line ending and no final newline, and expects lines 1, 3, 4 and 7. It also rejects a row loop that goes back to "index + 1". A new analyzer that copies the old parser now fails its own pull request.
Found by seeding bugs
This came out of a red-team run, where we plant small bugs in a file and see whether its tests notice. Of 24 planted bugs in the delivery commission analyzer, the tests caught 17 (71%). Two of the seven that got through pointed at real behaviour nobody had pinned: a file with no final newline, and a date written "6/3/2026" with a one-digit day. Both work, but nothing would have noticed if they stopped. Eight new tests later, all 24 are caught. (The tool re-checked 23 of them automatically; the parser fix rewrote the line of the last one, so it was re-planted by hand, and the new no-final-newline test fails on it.)
All 49 affected analyzers' test suites pass with the change. The code is MIT licensed, as always.