Fail-fast
Crash at the first sign something is wrong, rather than continuing in a state nobody understands any more.
The principle holds that errors should be detected as close to their cause as possible and produce an immediate, obvious stop. Concretely that means precondition checks at the top of functions, validating all configuration at startup rather than at first use, and never swallowing an exception because "it usually works out".
The motivation is debugging cost. A program that carries on with a broken invariant produces a symptom far from its cause, often as corrupted data discovered weeks later. Erlang's "let it crash" is the same principle taken to its conclusion: processes are made deliberately brittle and a supervisor restarts them, because a clean restart is more comprehensible than a partially damaged run.