Monitored Processes
dataset...
Mock pipelines
Outlier Days Excluded
warning0
Simulated samples
Operating Mode
verified_userDemo
Simulated data
Last Updated
update...
Sample snapshot
SLA Calculation Methodology & Statistical Learning Engine
How baseline performance targets and adaptive anomaly thresholds are modeled across pipelines
System Outlier Pruning
Evaluates 180-day telemetry. If a calendar day suffers a daily success rate < 50% (threshold = 0.50), it is tagged as an infrastructure/network crash and removed via an Anti-Join.
Temporal Sessionization
Partitions executions into 2-hour sliding buckets (hour // 2) to collapse retry storms into unified operational sessions with normalized diurnal start seconds.
Median Multiplier Filter
Computes the median duration (median_duration_s). Any hanging, zombie, or stuck sessions exceeding 300% (3×) of the median are purged from the quantile baseline.
Adaptive SLA Buffer
Extracts the 95th Percentile (P95) and adds an adaptive cushion of 15% (min 30 min) to set the critical operational SLA threshold.
T_expected = AvgStartTime + AvgDuration; Maximum SLA cutoff is T_max = AvgStartTime + SLA_Duration with automatic day-shift offsets (+1d).