After hours of analysis, Claude told a developer something his native recovery tools wouldn't: most of the data is gone.
The situation started when a forced reboot during kernel swapping corrupted a 12TB BTRFS array - BTRFS being a Linux file system format used in large storage setups for its snapshot and data integrity features. Standard recovery tools failed. The developer ran BTRFS's built-in repair utilities and got nowhere. He turned to Claude, walking it through error outputs, drive states, and file system structure across several hours of back-and-forth.
Claude's assessment was specific. The index table - the part of the file system that tracks where each piece of data physically lives on disk - was destroyed at roughly the 80% mark. Every node beyond that threshold was corrupted. The diagnosis: more than 8TB of data was unrecoverable. Not "may be difficult to recover." Gone.
The developer's account focused on how direct Claude was about it. AI tools have a reputation for hedging - for suggesting workarounds that don't exist, or softening bad news into something that sounds less permanent than it is. In this case, Claude identified the failure point precisely, explained why the damage was irreversible, and told him what percentage could still be salvaged.
The technical capability itself isn't surprising - BTRFS recovery processes are well-documented, and the analysis drew on publicly available file system documentation. What stands out is that Claude delivered a clear, unfavorable answer when that was the accurate one, rather than offering false optimism.
For anyone running large storage arrays: BTRFS snapshots are powerful, but they aren't a substitute for off-array backups. A corrupted index table can make the file system's own recovery tools useless.