A system can look redundant from the front and singular from behind.
Suppose five services have separate names, teams, and interfaces, but all authenticate through one provider. If that provider becomes unavailable, five services disappear together. The visible count was five. The number of working routes was one.
Standardization earns its place. It simplifies training, repair, compatibility, and oversight. But repeated designs can also repeat the same conditions of failure. Spare units help when individual parts break. They do not help when every unit shares the defect, supplier, controller, dataset, credential system, network, or power source that caused the break.
Under some tested conditions, bounded variation can make complex systems more stable because different components do not always fail under the same disturbance. But difference is not automatically protection. Different vendors may still depend on the same infrastructure. Different designs may become impossible to maintain together. Variation that destroys coordination creates another weakness.
For any function that must survive, trace each recovery route backward. Where do the routes merge? Which disturbance could reach all of them at once? Then test those junctions, not only the individual parts.
The useful arrangement may combine shared interfaces with genuinely separate implementations and recovery paths.
A backup is not defined by standing nearby. It is defined by what it can survive independently.