# Excel saved the file. Its total was still $1 behind.

Reproduce a native Excel manual-calculation failure: a budget input changes, the report stays stale, and an explicit rebuild restores the expected total.

Published 2026-11-15T07:30:00Z by Renaissance Superintelligence Corporation (RSI).

Check calculation mode and verify the expected result before accepting an Excel output. In our Panes experiment, changing a budget input from $140,000 to $140,001 left the report at $645,000 in manual mode. CalculateFullRebuild brought it to $645,001.

## A saved input and a stale report can coexist

We copied the synthetic expense workbook into a new task directory, opened it in native Excel and deliberately selected manual calculation. We also disabled calculation before saving for this test. The source budget table and the report were otherwise the same fixture used in the Power Query experiments.

The report began at $645,000. One October budget row contained $140,000. We increased that source value by $1 and read the report again: it still showed $645,000. Excel reported calculation state 2, pending. We saved a separate copy at that point so the stale state could be inspected.

This is a controlled demonstration of manual calculation, not evidence that Panes selects that setting by default. The script saves the previous settings and restores them in its cleanup. An agent should inspect the workbook and the Excel instance it actually received rather than assume a calculation policy.

## Rebuild, then test the financial change

The script called CalculateFullRebuild. Microsoft documents that operation as calculating all open workbooks and rebuilding formula dependencies. The report then returned $645,001 and calculation state 0, done. The source change and the report change reconciled exactly.

CalculationState describes Excel calculation activity. It does not prove that a query downloaded current records or that an external workbook was available. Our separate missing-source experiment reached a done state while still retaining old query data. Treat source freshness, formula calculation and financial reconciliation as separate acceptance checks.

For a task that changes formulas or inputs, write down the expected effect before performing the edit. Here it was a $1 increase in the total budget. That independent expectation caught the problem immediately; reading the saved file back without an expected value would merely have repeated the stale answer.

```powershell
$before = $report.Range('B9').Value2
# Apply the task's input change here.
$xl.CalculateFullRebuild()
$after = $report.Range('B9').Value2
if ([math]::Abs(($after - $before) - 1) -gt 0.005) {
    throw 'Budget total did not reflect the expected USD 1 change'
}
```

Observed native values in the controlled manual-mode test

Stage | Source budget row | Report budget | Calculation state

Before edit | $140,000 | $645,000 | Not recorded

After edit | $140,001 | $645,000 | 2: pending

After rebuild | $140,001 | $645,001 | 0: done

## Keep the calculation boundary visible

A successful Save is a persistence event. In this test it could preserve an unrecalculated result. Saving after the explicit rebuild gave us a different and correct output. That distinction matters when a later step reads cached cell values to build a presentation or a Word report.

Use an isolated Excel instance for this kind of test: a full rebuild affects every workbook open in that instance. Preserve and restore settings you intentionally change. For normal business work, select a calculation policy with the workbook owner rather than silently replacing a model’s existing manual mode.

The downloadable before-recalculation copy is deliberately stale. Reopening it under a different calculation policy may update it, so inspect its saved values before changing its state. Do not hand it to a customer as an accepted report.

## Download both states and the exact procedure

Inspect the [before-recalculation workbook](/research/native-office-2026-11-15/manual-before-recalculation.xlsx), [recalculated workbook](/research/native-office-2026-11-15/manual-calculation.xlsx), [result log](/research/native-office-2026-11-15/manual-calculation.json) and [PowerShell script](/research/native-office-2026-11-15/native-manual-calculation.ps1). The script expects expense-original.xlsx in a new directory; use the synthetic [expense workbook](/examples/excel-to-powerpoint/October%20close.xlsx) as that input.

Run the native step through [Panes desktop execution](/docs#commands). Pair this test with [query freshness](/blog/prove-excel-power-query-refreshed) and [pivot reconciliation](/blog/refresh-pivot-after-power-query) when the workbook contains all three layers. The useful acceptance condition is a reconciled output from current inputs.

## Evidence

Prepared scripted execution on one retained 4-vCPU, 16-GB Windows research computer, November 15, 2026. Excel 16.0 build 20527. Recalculation took 0.0440077 seconds; the native script reached its saved result in 3.003 seconds before finally-block cleanup, API time was 3.898 seconds and controller wall time 3.976 seconds. One observed run, not an autonomous-agent or fleet benchmark. Per-task operating cost is unmeasured. Public Office copies have author metadata redacted; all other ZIP members are identical to the native outputs. They were reopened read-only in Office without changing their bytes.

## Sources

- [Microsoft: CalculateFullRebuild](https://learn.microsoft.com/en-us/office/vba/api/excel.application.calculatefullrebuild)

- [Microsoft: CalculationState](https://learn.microsoft.com/en-us/office/vba/api/excel.application.calculationstate)

- [Exact native result](https://usepanes.com/research/native-office-2026-11-15/manual-calculation.json)

- [Public-file identity and native reopen](https://usepanes.com/research/native-office-2026-11-15/publication-manifest.json)