A workbook total represents the inputs and calculation state used to produce it at a particular business cutoff. When the workbook retrieves external data, refresh timing becomes part of that meaning. A mathematically correct formula can produce an unsuitable operational result if it uses stale or differently timed inputs.
This is a conditional failure mechanism to test in a workbook that combines external connections, formulas, VBA, and publication. It does not establish how a particular workbook is configured or how often stale results occur.
Refresh is an operation with a result
Excel stores connection information describing how to locate, authenticate to, and query an external source. Refresh can be configured on opening or at intervals. Those settings request retrieval under defined conditions; they do not demonstrate that retrieval succeeded.
Background refresh can return control before data retrieval finishes. Microsoft’s documented option excludes OLAP queries and connections retrieving data for the Data Model, so the actual connection type matters.
If later calculation or publication assumes that a refresh request means completion, the workflow can cross its acceptance boundary too early. The test should establish the completion condition used by the actual workbook and any controlling VBA.
Trace the total through time
Take a constructed cash report that combines a database extract with manual adjustments. Its result depends on the extract’s cutoff, the adjustments included, the formula version, and the time at which the report is published.
A replacement calculation using a later database snapshot can differ for a legitimate timing reason. Conversely, matching a previous total might indicate that the workbook reused previous inputs. Numerical equality alone cannot distinguish those cases.
Record source keys, business dates, refresh results, calculation steps, adjustments, and publication state for the comparison. These observations make it possible to explain a discrepancy rather than classifying every difference as arithmetic failure.
A connection failure can leave a plausible artifact
An unavailable database does not prevent the workbook file itself from existing or opening. The failure exercise should determine what values remain visible and whether the user can publish them without recognizing their age.
The ODBC path also needs inspection under the actual Excel architecture and execution identity. A Windows user DSN can be listed even when its driver cannot be used by that process.
Preserve credential references without copying secret values into the evidence record. Test both the successful query and the workflow’s response to failed authentication, interruption, or an unavailable source.
Reconciliation explains differences
US Treasury agency reconciliation guidance requires differences to be identified and explained, with adjustments traceable to supporting documents. Its federal accounting context does not define every corporate tolerance, but the distinction between agreement and explanation is useful here.
Matching keys, currency, fees, aggregation, timing, and rounding need treatment under the applicable business policy. An adjustment added solely to force totals to agree does not explain why the original difference arose.
Expected exceptions should have an owner and disposition. The evidence should distinguish a recognized timing difference from an unresolved missing transaction.
Compare the complete publication cycle
A governed workbook, an extracted calculation service, and a replacement application should be compared using matched starting inputs and business cutoffs. Any approved normalization must preserve financially meaningful differences.
The cycle includes refresh, calculation, macros, manual adjustments, save, and distribution. If the workbook writes back to a database, that transaction path requires its own authority and recovery tests.
Acceptance establishes which population and period the published total represents, how failed refresh prevents or qualifies publication, and how differences are resolved. The result is a reproducible business output whose timing can be explained alongside its arithmetic.