Definition
Coherence of system state across services after multi-service rollback events, measuring whether individually successful rollbacks compose into a consistent whole.
The Measurement Gap
Measurement Gap and Decision
The Gap: Traditional deployment metrics completely miss this failure mode. Deployment frequency is still high. Change failure rate looks fine because each individual rollback "succeeded." MTTR is fast because automation responded quickly. The gap is between individual rollback success and collective rollback coherence. You can have a system where every single automated rollback works perfectly in isolation yet the interaction between rollbacks creates more instability than the original failed deploy would have caused.
The Decision:
Are automated rollback policies across services producing coherent collective outcomes, or are locally correct rollback decisions creating system-level instability? This is a coherence metric, not a speed metric: it doesn't care how fast your rollback executes, it cares whether your rollback creates collateral damage in dependent services.
What Changes When You Have This Number:
Above 0.85, rollback policies are well-isolated and cascades are rare. Between 0.70–0.85, investigate specific service pairs involved. Below 0.70, cross-service rollback policies need review and deployment group coordination should be considered. Below 0.60, automated rollback may be actively creating more instability than it prevents in its current configuration.
Formula
Formula:
RCCI = Rollbacks_clean / Rollbacks_total
Where Rollbacks_clean is automated rollback events where no dependent service experienced a cascade effect (secondary rollback, unplanned scaling event, or error spike > 2x rolling 7-day baseline) within W minutes (30 recommended), and Rollbacks_total is all automated rollback events in the period.
Sign Convention:
Higher is better. 1.0 = no rollback ever triggered a cascade. 0.0 = every rollback propagated instability.
Worked Example:
Over Q3, 42 automated rollbacks occurred:
- 8 triggered secondary rollbacks in dependent services within 30 minutes
- 5 triggered unplanned scaling events in dependents (no secondary rollback but error-driven auto-scaling)
- 3 triggered error-rate spikes > 2x baseline in dependents (no rollback or scaling, but measurable degradation)
- 26 completed cleanly with no detectable downstream impact
RCCI = 26 / 42 = 0.619
This falls below the 0.70 threshold, indicating systemic cascade risk. Analysis of the 16 cascading events reveals that 11 involved the same three tightly-coupled service pairs, suggesting those pairs should be deployed and rolled back as coordinated groups rather than independently.
Exclusions:
Simultaneous planned rollbacks (intentional coordinated deployment groups) are excluded from the denominator. Those aren't cascades, they're designed collective responses.
Proposed formula, pending validation pilots.