Failure density describes the relationship between the amount of failure observed and the amount of change introduced into a system over a given period of time.
Rather than counting failures in isolation, failure density considers how much of the system was in motion when those failures occurred. This provides a more meaningful signal of system health than raw failure counts alone.
Failures are not evenly distributed across time or work. A system undergoing significant change will naturally encounter more risk than a system that is static. Comparing failure counts without accounting for change surface obscures this reality.
A single failure during a large, high-risk change may indicate resilience. Repeated failures during small, contained changes often indicate fragility.
Failure density shifts the question from:
- “Did something break?” to:
- “How much failure occurred relative to how much the system changed?”
Observed over time, failure density can surface patterns such as:
- Systems that absorb large change with minimal failure
- Areas of the system where small changes repeatedly lead to incidents
- Hidden coupling or architectural brittleness
- Improvements in resilience that are not visible through delivery metrics
Because it normalizes failure against change, failure density allows comparison across projects of very different sizes without reducing them to productivity measures.
Failure density is not:
- A quality score
- A performance metric
- A measure of individual or team effectiveness
It does not assign blame or imply that failure is unacceptable. Failure is an expected part of system evolution. Failure density exists to help reason about how tolerable that failure is in context.
Failure density is most meaningful when observed across multiple cycles of change. Single data points are rarely informative on their own.
Trends in failure density—rising, falling, or stabilizing—provide insight into whether a system is becoming more resilient, more fragile, or simply changing in risk profile. When paired with recovery time and context, it becomes a central signal in understanding system health.