If January's total appears in February or one category inherits another category's value, inspect the table-calculation settings before rewriting the data source. A plausible new formula can conceal a worksheet configuration problem.
Specify the small view you expect
This original fictional worksheet has Category on Rows, Month on Columns in chronological January–February order, and SUM(Units) on Text. There are no additional dimensions or filters in the fixture:
| Category | January units | February units |
|---|---|---|
| Books | 10 | 5 |
| Kits | 20 | 7 |
The task is cumulative units across months, restarting for each category. The expected result is Books: 10 then 15; Kits: 20 then 27. It is not a cumulative total across category rows.
The candidate table calculation is:
RUNNING_SUM(SUM([Units]))
These are authored fixture expectations, not a claim that Tableau or ChatGPT ran the workbook. Tableau's table-calculation documentation, checked October 24, 2026, explains that the calculation works over the view. Addressing determines its direction; partitioning separates the groups in which it runs.
Ask for a cell trace
Help diagnose a Tableau table calculation. Do not modify a workbook.
Reader question: cumulative units by month separately for each category.
Rows: Category, in Books then Kits order.
Columns: Month, January then February in chronological order.
Marks: Text = SUM(Units). No other dimensions or filters in this fixture.
Formula: RUNNING_SUM(SUM([Units])).
Values: Books Jan10 Feb5; Kits Jan20 Feb7.
Current Compute Using setting: [COPY ACTUAL SETTING OR UNKNOWN].
Explain addressing versus partitioning in terms of this worksheet.
Trace the expected four cells for the intended across-month result.
Then trace Table (down) so I can recognize the wrong direction.
Suggest the Compute Using setting to inspect, including Specific
Dimensions where appropriate. Preserve the source rows and formula
unless you identify a separately explained reason to change them.
Flag any actual worksheet detail I have not supplied. Do not claim
you opened, changed, or verified my workbook.
OpenAI's prompting guidance supports giving the information that changes the answer. Here that information includes the view, not merely the source CSV.
Recognize the wrong traversal
For the stated layout, Table (across) travels from January to February within each category row. Table (down) travels from Books to Kits within each month column. The wrong-direction fixture therefore becomes:
| Category | January, running down | February, running down |
|---|---|---|
| Books | 10 | 5 |
| Kits | 30 | 12 |
The 30 is 10+20, and 12 is 5+7. Those values can be mathematically correct for the wrong question. Do not change Kits' January source value to make 30 disappear.
Open Edit Table Calculation on the calculated field and inspect Compute Using. For this simple layout, start by checking Table (across). Under Specific Dimensions, the intended addressing dimension is Month and the partitioning dimension is Category. Record the actual selected fields and the order rather than relying on a screenshot of a different worksheet.
Recheck after the view changes
Sort order matters to a running total. An alphabetical month label can put February before January; a correctly directed calculation would then accumulate in that wrong sequence. Use the appropriate chronological date order and inspect the headers.
Additional dimensions on Detail, a changed layout or filters can change the calculation's visible domain. Repeat the cell trace when you add them. The four-cell fixture does not settle how your real workbook treats missing months or filtered records; supply those requirements separately.
Check the native worksheet for Books 10/15 and Kits 20/27, then preserve the formula, addressing, partitioning and month order with the workbook. If the next task is choosing a chart for those checked values, the data-visualization guide addresses axes and missing data separately.