When a primary-driven index build resumes after a failover, its collection scan seeks to exactly the smallest lastSpilledRecordId across the build's indexes. If that document was deleted before the resume, for example by capped collection roll-off or a user delete, the seek fails with KeyNotFound and the new primary aborts the build. The scan should instead start at the first remaining record at or after the resume point, which is safe because the deleted documents' keys are removed when the side writes table is drained.