DuckDB + dbplyr: When Your Pipeline Gives Different Results Every Time It Runs

DuckDB parallelizes query execution and never guarantees row order unless explicitly specified. When using DuckDB with dbplyr in R, four common patterns silently produce non-deterministic results: window functions (row_number, cumsum, lag) without explicit window_order(), distinct(.keep_all=TRUE) which picks rows arbitrarily, inequality joins that create fan-out when reference periods overlap, and synthetic row expansions that discard the expansion index. Each pattern is explained with concrete examples and deterministic fixes. A code review checklist and a method to detect residual non-determinism by running the pipeline multiple times are also provided.