This solution does not reliably meet the requirement. The underlying problem is not the absolute count of rows in the CSV but how Power Query (Get & Transform) infers and expands JSON properties: Power Query samples the input rows to discover the JSON schema/fields. If a particular JSON property does not appear in the sample used for schema detection, Power Query will not create a column for it when you expand the JSON. Simply reducing the exported record count from Defender (so you end up with fewer rows) only helps if the exported dataset by chance contains the missing JSON properties within the rows Power Query samples. It is not a guaranteed or controlled fix and therefore does not meet the requirement.
The correct fixes are to (a) adjust Power Query's sampling/preview settings so it inspects more rows (so the property is included in schema detection), or (b) explicitly parse the JSON and expand fields using Power Query M functions (for example parse each AuditData with JSON.Document and then expand Record fields), which ensures the desired properties become columns regardless of initial sampling. Reducing Defender export size is not the recommended or reliable approach.