[SuruchBoss]

What this changes

Closes #241.

The bug: ExcelJS hands a date-formatted cell over as a JS Date, and rawForDate in src/lib/excelIO.ts turned it into ISO text with isoFromSerial. For a serial of 0 or below, that text is a year-1899 day (1899-12-31, 1899-12-30, 1899-12-29 12:00). dateLiteral refuses any year before 1900, so the cell kept the string as text. As a result:

• SUM skipped those cells without a word;
• =A2+2 gave #VALUE!;
• export wrote the text over the number in the file.

The fix: rawForDate keeps a serial below 0, or a 0 that is not midnight under a time-only format, as the number itself. The cell's format comes in exactly as before, so the grid shows it the way #219 does: 00/01/1900 for 0, and ######## for a negative. Export writes the number back with its format. Midnight under a time-only format still opens as 00:00, the case the issue notes is already fine.

Reproduced on main first

The issue's file: A1 0 (dd/mm/yyyy), A2 -1 (dd/mm/yyyy), A3 -1.5 ([h]:mm), A4 1.5 ([h]:mm), A5 45366 (dd/mm/yyyy), B2 =A2+2, A8 =SUM(A2:A4) ([h]:mm).

	main	this PR
A1, A2, A3 hold	1899-12-31, 1899-12-30, 1899-12-29 12:00 (text)	0, -1, -1.5 (numbers)
A1 / A2 / A3 shown	the ISO text	00/01/1900 / ######## / ########
=A2+2	#VALUE!	1
=SUM(A2:A4)	36:00, i.e. 1.5 (A2 and A3 skipped), per the issue	-1
Exported A1:A3	the strings	the numbers, with dd/mm/yyyy / [h]:mm
A4 1.5 and A5 45366	36:00, 15/03/2024	unchanged

How it was verified

• npm run verify (all ten gates): all ten gates green on 5c1136b. The current head 8cc1b64 carries that tree unchanged; it only adds the merge of main 726af92 (Point each exported chart series at its own column (#59) #240).

Gate	Result
lint · check:readme · check:screens	pass
check:deps	every advisory fixed or accounted for
test	2028 passed (2017 by the README's static count)
check:mutants	32/32 killed
build · check:bundle	3804/4200 KB, largest chunk 934/1300 KB
check:a11y	121 checks passed
check:e2e	60 flows passed

• excelIO.test.ts (+3), on a real .xlsx written with ExcelJS in the issue's layout:
  • the cells open as the numbers, shown as Excel shows them, with the positive date and time unchanged;
  • formulas read them: =A2+2 is 1 and the SUM is -1;
  • export writes the numbers back with their formats, read back with ExcelJS, and A2 is not a string cell.
• Re-checked against the old code: with main's excelIO.ts, all three fail. The first finds 1899-12-31 and 1899-12-30 where the numbers should be, and the second gets #VALUE! for =A2+2.
• Not tried by eye on a production build. The tests drive the real importer and exporter on a real .xlsx, but I did not open the file in the app.
• READMEs, both languages:
  • a sentence after the zero-and-negative paragraph of เซลล์รูปแบบเวลาสะสม [h]:mm โชว์ชั่วโมงของวันแทนชั่วโมงรวม · Elapsed-time formats like [h]:mm show the hour of the day #219, saying an opened file now does the same;
  • the excelIO.test.ts row (58 → 61).
• Recounted: 2017 tests in 145 files. Picture 27 retaken in both languages and checked by eye, blessed after the retake. Social card rebuilt and checked by eye.

The branch

This is the session's branch claude/new-session-m6mehm. The work was written on a local branch on top of #59's head 6f54079, so the counts already included #59's test, with main 726af92 (#240) merged in.

🤖 Generated with Claude Code

https://claude.ai/code/session_013pmWZsbmugfpqtYazrzfjj

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
excel-to-go	Ready	Preview	Oct 5, 2026 5:21am UTC

[SuruchBoss]

PO review: merged at 8cc1b64 (merge commit 448a5b2).

• npm run verify locally on 8cc1b64: all ten gates green. 2028 tests passed, mutants 32/32, a11y 121 checks, e2e 60 flows.
• CI: 17 checks green, CodeQL included.
• Tried in the app, since the PR said it had not been opened there yet. I used a production build at 1280 and 390 with the same .xlsx that found Opening an .xlsx turns 0 and negative date/time cells into text ("1899-12-31"), and export writes the text back #241. All of it now matches what Microsoft Excel shows:
  • 0 under dd/mm/yyyy shows 00/01/1900.
  • -1 under dd/mm/yyyy and -1.5 under [h]:mm show ########.
  • =SUM(A2:A4) under [h]:mm is −1, so it shows ########. Before this PR it was 36:00 because the text cells were skipped.
  • The formula bar on A2 reads -1.
  • =A2+2 is 1. IFERROR(TEXT(A2,"dd/mm/yyyy"),…) takes the fallback.
  • Typing 5 into A2 shows 05/01/1900, and the total goes to 120:00. Undo puts back ########, and typing after the undo lands where it should.
  • Positive values are unchanged: 1.5 → 36:00, 45366 → 15/03/2024, and 0.5 under date+time → 00/01/1900 12:00.
  • Export writes the numbers 0, −1 and −1.5 back with dd/mm/yyyy, dd/mm/yyyy and [h]:mm, read back with ExcelJS.
  • No page errors, alerts or overflow at 390.

This was the last status:sprint-final ticket in the hold set.

---

Generated by Claude Code