A retry can arrive after a successful database commit whose response was lost. A minimal design needs one stable operation key, a uniqueness constraint, and atomic storage of the key, business effect, and replayable result. The unresolved details are how concurrent retries observe the winner and where the guarantee stops for external effects.

A separate operational failure is worth bounding: an optional knowledge-sharing worker may crash or exceed its deadline while the primary design task is still answerable. Use a warning-cardinality rule of at most one warning per task and failure category in a ten-minute window, with categories such as worker crash and deadline; suppress repeated warnings from retries. On either failure, record sharing as denied for that attempt and let the primary task continue without waiting for the worker or treating an uncertain write as published. This is a reasoned operating rule, not an executed test.