Background
When RealTheory generates right-sizing recommendations, it analyzes your Kubernetes environment in depth. For workload recommendations, this centers on CPU and memory usage patterns over time. For node pool recommendations, the analysis goes significantly further — translating raw resource demand into real node utilization across overhead, scheduling, fragmentation, packing efficiency, and cost. Recommendation Confidence reflects how much observed data was available to inform that analysis.
More available data results in higher confidence. Less available data results in lower confidence.
Why confidence matters
The confidence level helps you interpret a recommendation with the right level of caution or trust. It is a signal — not a verdict — that blends RealTheory's automated analysis with your operational insight.
- RealTheory brings the statistical view.
- You bring the resource knowledge.
Examples
Example 1: A stable, predictable workload or node pool
Suppose RealTheory has only 2–3 days of data for a workload or node pool with steady, uniform usage — for example, a background job with minimal variance, or a node pool running a consistent mix of production workloads with predictable pod density and no major scaling events.
Even if confidence is reported as Low or Medium due to the smaller sample size, the recommendation might still be trustworthy because the behavior is consistent and representative. Your knowledge of the workload or node pool adds valuable context to the reported confidence level.
Example 2: A spikey or variable workload or node pool
Now consider a workload or node pool that experiences bursty, unpredictable demand — for example, an on-demand API with variable traffic, or a node pool dedicated to batch jobs that run on a schedule, leaving nodes largely idle between runs.
If RealTheory has only 24 hours of data, there is a good chance those spikes or idle periods were not fully captured. The resulting recommendation might underestimate actual resource needs, and confidence will be Low — a signal that it is best to wait for more data before taking action.
How to use confidence effectively
| Confidence level | Meaning | Guidance |
|---|---|---|
| High | Sufficient data for reliable analysis | Safe to act |
| Medium | Some data is available, but not yet ideal | Generally reliable — verify against known behavior patterns |
| Low | Limited data available | Use caution — verify against known behavior patterns and consider waiting for more metrics |
Best practices
- Don't ignore low confidence, but don't fear it either — it's just a signal.
- Review behavior alongside confidence to make informed decisions.
- Use RealTheory recommendations iteratively — apply, observe, refine.