Propagation depth

Propagation depth is the number of successive hops a fault survives through an agent system before a check contains it. Hops are the unit that matters because each hand-off is a containment opportunity: a fault that lives through k hand-offs has passed k gates that could have stopped it, and every stage that builds on the corrupted result compounds it further. Depth is the hop count behind error propagation, and it pairs with propagation radius, which reads the same fault’s breadth, how many agents it touches, rather than how far it travels.

Depth is read per topology. In a chain of depth D, an uncaught fault at step k can survive every remaining hop, so depth tracks the length of the pipeline. In an orchestrator fan-out, depth stays shallow while breadth carries the exposure, since one bad return reaches many workers in a single hop. Report it per injection point as a distribution, never one headline number.