A query with no visible errors can still be incomplete if the failing rows were removed. Keep enough evidence to explain which records were accepted and which need attention before describing the import as ready.

Make a small Power Query practice source with four identifiers, R1 through R4. Give their text amounts the values 12, 7, pending and 5. Retain an untouched RawAmount column alongside the Amount column that will be converted. This makes the rejected text available even after the conversion result becomes an error.

Separate a failed step from a failed cell

Microsoft distinguishes step-level failures, which prevent the query from loading, from cell-level errors inside a table. Selecting whitespace in an error cell shows its details. Keep Errors works on rows with errors in the selected column; it is not a recovery command for a missing source or a step that cannot produce a table.

Convert only the practice Amount column to a number. Preserve two copies of that conversion result. In the exception copy, select Amount, then Home, Keep Rows, Keep Errors. In the accepted copy, select Amount, then Home, Remove Rows, Remove Errors. Keep the same column scope in both operations.

The accepted identifiers should be R1, R2 and R4, totalling 24. The exception set should contain R3 with RawAmount still showing pending. Three accepted records plus one exception account for the four inputs; neither set should contain an identifier from the other.

Reconcile identity as well as the count

Compare the combined identifier sets with the original source. A matching total count could hide one duplicated record and one omitted record, so inspect the actual identifiers. Keep the reason for R3's rejection with the exception list rather than replacing it with zero to make the total look complete.

Decide how the source owner should resolve pending: supply a confirmed amount, retain it as unresolved, or apply an agreed exclusion rule. The query cannot infer that business decision from a conversion failure.

After the source is corrected, rerun both outputs and reconcile again. The previously rejected identifier should have an explainable new state. For a larger import, preserve the exception output with the reporting period and conversion step so a clean-looking accepted table cannot conceal discarded records.

Sources: Microsoft: query errors.