schelk

Run `npx skills add tempoxyz/schelk --all -g -y` in your terminal to install the skill globally. The schelk binary itself is built from source on a Linux host with `cargo install --path .`.

It restores large benchmark datasets, like database or reth datadirs, to a clean baseline in seconds instead of hours. It does this without the measurement distortion caused by ZFS, btrfs, or LVM overlays.

It tracks exactly which blocks were written using dm-era metadata on a ramdisk, then copies only those changed blocks back from the pristine virgin volume. Recovery time depends on how much was written, not the total volume size.

You need a Linux machine with two equal-size block devices (one virgin, one scratch) plus a ramdisk for metadata, typically 4 GiB for a 1.7 TiB drive. All operations require root privileges.

Ramdisk metadata is lost on reboot, so incremental recovery is no longer safe. Run `schelk full-recover` to copy the entire virgin volume back to scratch before continuing.