Datadog APM and Infrastructure Monitoring: Adaptive Sampling Impact on Bottleneck Visibility

Goal: Determine how Datadog's adaptive sampling algorithm decides which traces to retain or discard so that bottleneck analysis remains reliable before optimization.

Constraint: Public documentation describes adaptive sampling as traffic- and error-driven but does not disclose the exact weighting of latency, error, and throughput metrics, creating uncertainty about whether rare high-latency spikes are preserved when overall volume is high.

Questions: What specific latency, error, and throughput thresholds does the algorithm use to adjust the sample rate? How does it prioritize retaining spans with elevated latency versus those with errors or high throughput? Can users configure or influence the weighting to favor visibility of infrequent but high-impact bottlenecks?