Adequacy of the ACR=0 criterion for detecting performance degradation

Ascertain whether defining the onset of the Performance Degradation State in Online Collaborative AI Systems (OL‑CAIS) by a continuous drop of the Autonomous Classification Ratio (ACR) to zero is an adequate criterion, or whether this detection rule should be refined to more reliably capture performance degradation across diverse scenarios.

Background

The resilience model introduced in the thesis uses the Autonomous Classification Ratio (ACR) to track OL‑CAIS performance, with a continuous decrease to zero indicating entry into the Performance Degradation State. During experiments, instances arose where degradation was present but not detected by this rule, suggesting potential shortcomings of the current criterion.

Clarifying the adequacy of this thresholding approach—and possibly refining it—would improve automated detection of disruptive events and strengthen the reliability of resilience assessments in OL‑CAIS.