The practical answer
In this example, 2024-01-01 is 45292 in the 1900 system and 43830 in the 1904 system. Add 1462 when moving a confirmed 1904 serial to a 1900 target, or subtract it in the reverse direction. Keep the raw value, source system and a known calendar date as controls.
Distinguish three things that look like dates
A displayed 2024-01-01 might be a numeric serial with a date format or simply text. Matching appearances do not establish matching meanings. Inspect a copy in General format without overwriting the original, and record the workbook’s date-system setting.
Microsoft’s date-system documentation establishes the 1,462-day difference and gives Windows and Mac settings paths. The computer brand is not evidence of a particular file’s setting: older templates, exporters and manual settings also matter.
Formatting changes appearance; it does not identify the source convention. A lone 43830 in a CSV, without a system or known date, could be an ID rather than 2024-01-01. Ask the source owner and retain an unresolved state instead of guessing.
A hand-checkable delivery ledger
These are original teaching records, not customer orders or an executed Microsoft Excel test. Two stores report the same day; a third extract is already converted, while the fourth explicitly contains date text. None represents a timezone timestamp.
| Record | Source value and type | Confirmed state | Target |
|---|---|---|---|
| A | 45292, numeric date | 1900 system | 45292 |
| B | 43830, numeric date | 1904, not converted | 45292 |
| C | 45292, numeric date | Already converted from 1904 | 45292, no extra adjustment |
| D | 2024-01-01, text | Explicit calendar date | Keep text or parse separately under target rules |
Convert only confirmed numeric dates, once
In a delivery copy, create raw_serial, source_system and target_serial columns. Copy A unchanged; calculate 43830+1462=45292 for B; keep C at 45292. Carry the conversion state with the result so the next recipient does not repeat it.
Check the reverse direction too: 45292−1462=43830 when a 1900-system value is deliberately prepared for a 1904-system recipient. Direction follows source and target conventions, not a visual guess that a date seems early or late.
Do not switch an entire workbook’s system blindly to repair a few imported rows. That changes the interpretation of other date values. A copy with an explicit affected-field conversion is easier to review. Preserve fractional parts in date-time serials rather than rounding away the time.
Detect a second conversion
C: 45292 + 1462 = 46754
C: retain 45292; mark already-normalized
Check a second known date and the unaffected fields
Add 2024-01-02 as a neighboring control: serial 45293 in the 1900 system and 43831 in the 1904 system. The two representations must differ by 1462, while neighboring dates remain one day apart. This also catches a manual row shift.
Reconcile record IDs and row counts. Leave non-date amounts and IDs unchanged; do not replace blanks with an epoch date or add a number to explicit date text. Save, reopen and inspect the final artifact in the receiving software.
For a text handoff, declare YYYY-MM-DD and the intended field type. This is a calendar date, not midnight in an implied timezone. The recipient must still decide whether to import it as text or a date. Text export does not resolve an unknown source serial.
Acceptance conditions
- Every source has a recorded system and at least one known calendar date.
- Only confirmed numeric dates are converted; text, blanks and IDs are excluded.
- A/B/C normalize to 45292; the next day normalizes to 45293.
- Keep the original, corrected copy, field types and conversion record traceable.
Merge only after date meanings agree
Then use the different-header merge method to align fields. Successfully appending rows does not prove that date meanings agree; identically named columns can contain serials from different systems.
For CSV delivery, also read the cached-formula and multi-sheet export guide. CSV does not carry the full workbook setting, so include source and field documentation. This article does not claim automatic correction by every Excel paste path or by KitNelo converters.
Common questions
Does any four-year difference prove a date-system fault?
No. An exact 1,462-day difference plus matching source settings, known dates and conversion state is stronger evidence. Mistyped years, two-digit year interpretation and text parsing are separate possible causes.
