Case study / 01
Safe Artifact Lifecycle Automation
A staged automation system that inventories, classifies, plans, quarantines, and only permits destructive cleanup after fresh evidence and explicit approval.
The problem
Long-lived automation estates accumulate archives, intermediate artifacts, duplicates, and files whose producer is no longer obvious. Cleanup has to reduce clutter without treating age, size, or unknown ownership as permission to delete.
The system
DRH built the 00.07.a–f lifecycle chain around discovery first: census and producer mapping, lifecycle classification, cleanup planning, reversible quarantine, delayed destructive cleanup, and read-only storage reporting. Unknown artifacts stay protected, duplicate decisions require evidence, and destructive actions remain separately gated.
The failure that mattered
During the first real duplicate quarantine, the workflow reported success after moving a 1.16 GB archive, but a later storage census showed an unexpected drop. Investigation proved that the quarantined copy survived while the intended canonical retained copy was missing. The success condition had verified what moved, but not the thing that was supposed to remain.
The correction
The canonical archive was restored from quarantine and verified by size and SHA-256. The quarantine executor was then patched so it cannot report success unless the retained canonical archive and checksum evidence still exist and match the approved state after the mutation.
What it demonstrates
The resulting lane favors reversible actions, explicit state, accounting invariants, postcondition verification, and human approval before destructive cleanup. It is now scheduled for weekly census, classification, planning, and reporting, while quarantine and deletion remain manual.