The Actual Root Cause of the Missing-Rows Bug

The correct answer is "A single-column cursor with no tiebreaker skips same-timestamp rows."

The export's keyset pagination cursor advances on created_at alone with a strict > comparison. If a row is inserted with the exact same timestamp as the last-read cursor value, after the page containing that cursor value has already been read but before the next page is fetched, the strict > comparison skips it — the cursor has already advanced past that instant. It's a missing tiebreaker column on the pagination cursor, not a filtering bug.

The distractors describe things ruled out or never true in the case study: the timezone-boundary filter was the earlier, discarded hypothesis; nothing in the narrative involves the generated Prisma client caching stale results; and no rate limit is ever implicated in the missing rows.