Filtered spreadsheet updates hit the wrong rows: how do you locate records by stable ID?

Preserve target IDs and the actual write range

This applies to filtered bulk updates, stale ranges after insertion and asynchronous results written in arrival order. Preserve write ranges and source IDs and stop expanding the damage. Some spreadsheet APIs expose positional indices, not stable record identifiers. The application needs traceable business identity rather than treating every row object as a database key.

Separate visible row numbers, physical positions and business identity

Check whether reads include hidden or filtered rows and whether post-sort mappings still use an old snapshot. Inspect indexing bases and addresses rather than equate first visible row with first physical row. Check concurrent movement; without version conditions, consider serializing critical updates instead of writing stale positions.

Relocate before writing and inspect concurrent changes

Use an ID column or explicitly supported metadata association and locate a unique current target before checking key fields. Stop on duplicate or missing IDs rather than choose a nearby row. Update only required fields and preserve formulas. Where needed, compare current and previously read values before submission instead of overwriting entire rows from stale snapshots.

Read back targets after sorting, filtering and inserted rows

In synthetic sheets, sort, filter and insert rows before updating the same ID, then verify only the target changed. Include duplicates and concurrent edits and require explicit conflicts instead of silently selecting the first row. Acceptance means correct business identity and preserved neighbors. Correct counts and successful API status alone do not rule out wrong-row writes.

This article wrote no private sheets or undid edits. Metadata and row semantics vary by API; stable row IDs are not universal. Historical recovery requires business evidence, not resetting an entire sheet for one error. Keep uncertain targets in error rather than continue blind bulk updates.