Evidence-first diagnosis and recommended verification steps. Deterministic facts stay separate from hypotheses, with optional local LLM enrichment.
Your largest immediate opportunity is recommendation_generation. The primary signal moved 154.9% above baseline while median runtime increased 115.1%, making this the first workload to investigate.
How LakeOps keeps recommendations grounded
Observed spend is materially above the recent baseline. Common causes are longer runtime, higher run frequency, a compute/SKU change, or increased input volume.
Review recent runtime, schedule, compute, or workload changes for this job.
This workload represents a large share of observed spend, so even modest optimization can have meaningful ROI.
Prioritize this workload for rightsizing and schedule/runtime optimization.
A meaningful share of recent runs ended unsuccessfully. Compare the first failing run with prior successful runs to isolate schema, dependency, credential, or environment changes.
Inspect recent failed runs and upstream/schema/environment changes.
Repeated retries suggest unstable dependencies, transient infrastructure failures, or retry-prone task logic.
Identify transient dependency failures or unstable task logic causing retries.