Owned public project · rudycelekli/gradia-range
gradia-range · Replayable adversarial tasks in synthetic twins
Problem
Testing a risky agent workflow against a live system can expose operational data and make outcomes hard to reproduce.
Design and implementation
Gradia Range describes cloning a system into a synthetic twin, slicing it into witnessed tasks, and producing replayable verdicts. Humans and agents attempt tasks on separate boards.
Evidence and scope
The source documents the evaluation design and release timing. Synthetic coverage depends on how faithfully the chosen twin represents the workflow.