The real issue here isn't a formula problem, it's a data architecture problem dressed up in spreadsheet syntax. A dedicated field biologist is spending hours wrestling `COUNTIFS` and `FILTER` functions because Survey123 treats a corrected entry as a new record rather than an update to the original. That's not a user error; it's a design failure that forces domain experts to become accidental database administrators.
What this means in practice is that every time a salmon gets misidentified, Coho instead of Steelhead, the correction workflow breaks the entire counting system. The app exports two sheets: one with individual fish measurements (including the errors) and another with totals computed from those errors. Changing the original record in the app doesn't retroactively fix the totals. So the biologist must export both sheets, manually reconcile timestamps that don't match between the fish-level data and the site-level data, and then rebuild the counts from scratch in Excel. That's not productivity; that's penalty labor for using a tool that can't handle the most basic data integrity requirement: allowing corrections to propagate.
The attempted `ROWS(FILTER(...))` formula reveals the deeper friction. It tries to match fish timestamps to site timestamps, but those timestamps are generated at different points in the workflow, one when the fish is measured, one when the site visit is logged, so they rarely align cleanly. Even if the formula syntax were perfect, the underlying data model fights it. The biologist has already fixed the individual fish bins with `COUNTIFS`, but the aggregation step remains broken because the two sheets were never designed to talk to each other after export.
The lesson is straightforward: if your data collection tool can't handle corrections without manual spreadsheet surgery, it's not a tool, it's a tax on your time. Field teams shouldn't need to become Excel power users just to get accurate fish counts. The solution isn't a better formula; it's a system where edits flow through to totals automatically, timestamps are consistent across related records, and the export produces one coherent dataset instead of two mismatched ones. Until that exists, the spreadsheet workaround is necessary, but it's also a clear signal that the current workflow is due for replacement, not repair.