[17:38:43] step 1: 2G image, mkfs -n 4096, loop + dm-delay (read=15ms write=5ms)
[17:38:46] step 1: populate S1 with 100000 files, snapshot -r to S2 (S2 now shares every S1 block)
[17:39:08] S1 id=256 root level=3 (want >=2; this is the shape from IDEA.md step 1)
[17:39:08] step 2: delete S1 (S2 keeps every one of S1's root children referenced)
[17:39:08] step 3: sync-poll for a non-zero drop key at root level 3, then unmount
[17:39:14] poll 2: drop key (36835 ...) level 3 -- child dropped, unmounting
[17:39:15] step 4: resume/interrupt cycles, looking for drop key to go non-zero -> (0 0)
[17:39:16] attempt 1: drop key (0 ...) level 0
[17:39:16] BUG REPRODUCED: drop key went from (36835 ...) to (0 0) while S1's root item still exists
[17:39:16] step 5: uninterrupted mount, wait up to 300s for completion or abort
[17:39:56] abort signature in dmesg after 26s
[17:39:58] RESULT: kernel abort matches the incident signature (full log: ./work-repro-simple-unpatched-1/dmesg-step5.log)
[17:40:31] btrfs check --readonly rc=1 (full log: ./work-repro-simple-unpatched-1/check.log)
[17:40:31] RESULT: check reports mismatches matching the incident
[17:40:31] S2 content: DIFFERS or unreadable (100101 entries; ./work-repro-simple-unpatched-1/s2.manifest)
[17:40:31] summary: zeroed=1 aborted=1 check_bad=1 drop_done=0 s2_ok=0
[17:40:31] VERDICT: bug reproduced