During connection startup recovery, replaying a force-stop incremental backup log record clears the WT_BLKINCR structure using WT_CLEAR in __txn_system_op_apply (src/txn/txn_recover.c line 190) without freeing the previously allocated backup ID string (id_str). This leaks the id_str memory allocated by a prior backup registration log record replayed in the same recovery pass via __wt_backup_set_blkincr (src/cursor/cur_backup.c line 75). To deterministically reproduce this leak in a test without spawning flaky external subprocesses or causing environment-specific file pollution on CI (Evergreen), the test reconfigures timing_stress_for_test=[checkpoint_slow] and executes the force-stop cursor close in a background thread. This allows to immediately copy the database home directory to simulate a crash before the slow checkpoint completes. Subsequent recovery is then executed on the copied directory in the main process to replay both backup ID records in a single recovery pass, exposing and validating the leak fix. I added before/after outputs on JIRA.