Standardize how my app represents empty values. Find where the same conceptual field is stored inconsistently: sometimes null, sometimes empty string, sometimes the key missing entirely. Profile fields, optional settings, and form-submitted values are the usual offenders. Pick one convention per field type and enforce it: optional text fields are null when unset (not ''); arrays are empty arrays, never null, where the app expects to iterate; booleans are true/false with null only if 'unknown' is a real third state; numbers are null when unset, never 0 unless 0 is the true value. Normalize existing rows with a migration, normalize at the write boundary going forward (form submissions are where empty strings sneak in), and update reads to rely on the convention instead of defensive checks scattered everywhere. Don't change what values mean; a user's intentionally-empty bio and an unset bio can stay distinct if the product treats them differently, but then document the distinction in the schema. Don't blanket-convert 0 to null in numeric fields; 0 is often real data. Deliver: the convention per field type, the normalization migration, the write-boundary normalization, and verification: write and read each optional field in its empty state and confirm one consistent representation.