The repair just isn’t one other bigger dashboard. It’s a cross-layer proof document that joins the mannequin’s motion, the harness’s consequence and the system’s noticed state. Every automated change ought to carry the unique goal, the proposed goal, the proof used to make the substitution, the boldness rating, the ensuing assertion and an express human-review standing.
For an analysis or procurement dialog, I’d ask 5 questions. First, can the device distinguish a profitable restore from a profitable execution in opposition to the incorrect goal? Second, does it measure false-heals on adversarially perturbed locators fairly than solely measuring whether or not a check reruns? Third, can an engineer reproduce the choice from an audit document? Fourth, does the device abstain when proof is weak, or does it optimize for a inexperienced construct? Fifth, can its occasions be correlated with the applying’s runtime traces and the mannequin’s device calls?
I’d additionally require a staged working mode. Autonomous restore can suggest a change, however high-impact adjustments ought to enter assisted triage till the group has proof that the restore preserves which means. That is related in spirit to chaos engineering’s emphasis on disciplined, observable experiments: the system ought to reveal its failure modes beneath managed situations earlier than it’s trusted in an uncontrolled one.
