Missing RCA and Mitigation task for some problems

[Tanaz-R]

Hello,
In this link https://github.com/microsoft/AIOpsLab/tree/main/aiopslab/orchestrator/problems, there are some problems like "ad_service_high_cpu", "network_loss", and some more that do not have "Analysis" and "Mitigation" tasks.
In your cited paper, the authors mentioned that "all functional faults are used to create problems at all of the
four task levels; while the symptomatic faults can only be used to create problems at the detection and localization levels".

Could you please explain the reason? Do you mean "Agent" will not be able to find them or fix, or AIOpsLab benchmark structure will not cover them or evaluate them?

I have another question as well, considering the "Mitigation" task, can you explain the evaluation, if the agent submits "Submit()" will pass, or if the agent submits and the whole state of the cluster is healthy, then it evaluates as "Success"?

Thanks for your support :)

[vitrixLab]

Hi @Tanaz-R ,

In AIOpsLab, functional faults are used across all four task levels, while symptomatic faults are limited to detection and localization because the benchmark does not define or evaluate RCA/mitigation ground truth for them. This is a benchmark scope decision, not an agent capability limitation.

For mitigation, success is evaluated by whether the cluster returns to a healthy state, not simply by calling Submit().

For more details, please refer to:
#131

Thanks.

[vitrixLab]

This issue helps understand how AIOpsLab evaluates success and mitigation (health-based, checkpoint-driven), but the limitation on symptomatic faults is a benchmark scoping decision, not a CI or agent constraint.