Metrics

Every run records a ground truth file with the exact times the fault was switched on and off, so each breaker decision can be judged against what really happened downstream.

Per-run metrics

Metric	Definition	Better
False trip rate (FTR)	Transitions to Open while the downstream success rate was above 95%, divided by all transitions to Open	Lower
Delayed detection (DD)	Time of the first transition to Open minus the time the fault started	Lower
Latency	Average, P50, P95 and P99 of client-observed latency	Lower
Resource overhead	Average and peak CPU % and memory (MB) per container	Lower

FTR = opens while downstream success rate > 95%  /  all opens
DD  = t(first transition to Open) − t(fault start)

A run without a fault, or where the breaker never opened, has no DD.

Statistical comparison

Per scenario, the fluxgo and gobreaker runs are compared with:

Step	Method
Normality	Shapiro-Wilk
Difference	Mann-Whitney U, two-sided, α = 0.05
Effect size	Cliff’s δ: negligible < 0.147 ≤ small < 0.33 ≤ medium < 0.474 ≤ large
Uncertainty	95% confidence interval of the mean (Student’s t)

A negative Cliff’s δ means fluxgo’s values tend to be lower, which is better for every metric above.

Micro-benchmarks

Per-call overhead of the library itself is measured with Go benchmarks; see Performance.