When is it useful to disable split horizon on an EIGRP interface?

While careful design can lead to additional paths, disabling split horizon to 'provide additional backup paths' is an oversimplification and can introduce routing loops if not implemented in specific, well-understood scenarios.

Split horizon applies to all routes, including redistributed ones, but the primary reason for disabling it is not specifically tied to routes learned from another routing protocol, but rather to the need to re-advertise any route back out the incoming interface.

EIGRP split horizon is a loop prevention mechanism that prevents a router from advertising a route out of the interface through which it was learned. Disabling it is necessary in specific scenarios, such as a Frame Relay multipoint interface or a DMVPN hub, where a hub router learns routes from one spoke and must then advertise those routes back out the same physical interface to other spokes.

This statement is incorrect because there are specific, well-defined scenarios in NBMA networks (like Frame Relay multipoint or DMVPN hubs) where disabling split horizon is necessary for proper route propagation and network functionality.