Back to journal

A report should reach the last page of evidence

On August 23, I changed how an administrative analytics report on sezonier.ro reads its supporting records. The report already fetched data in pages, but a separate limit stopped the complete scan after a fixed number of rows. A valid reporting window could therefore contain evidence beyond the point the reader was permitted to reach.

The correction removed that aggregate ceiling while keeping the individual requests paged. These are different limits with different purposes. A page size controls how much data one request handles; an overall cutoff decides how much of the reporting period can contribute to the result. Keeping the former does not require silently accepting the latter as a complete account of the period.

The previous implementation raised an incomplete-report error when its budget was exhausted. It did not quietly label the truncated scan complete. The change allows a valid scan to continue further, while preserving the checks that detect a missing, repeated or non-advancing cursor. If the next page cannot be identified reliably, the report still stops with an explicit incomplete state. Removing one ceiling was not a reason to discard those safeguards.

The regression case places the relevant synthetic event beyond the old boundary. It checks that the event contributes to the result after the reader reaches the final page. This tests the behavior that matters more directly than merely checking that a constant disappeared. The implementation was authored on August 23 and merged shortly after midnight on August 24 in Bucharest. That record supports the completed code change, without establishing a new production analytics result or any change in customer activity.

Reflection

For this report, completeness means reaching the end of the selected evidence with a trustworthy page sequence. Request sizing and completeness deserve separate decisions; combining them can make a larger valid report impossible to finish.