When to "let it crash" and when to defend the code in Erlang? -

so, "let crash" mantra erlang code meant resistant cruel world events unexpected pulling plug, hardware failure, , unstable network connections.

on other hand, there defensive programming.

being new erlang, wonder, how know when want process crash , when want defend flow if, case..of, type guards?

say, have authentication module, can return true/false result if authenticated or not. should have successful scenario , crash if user authentication fails due wrong login/password?

what other scenarios, like, if product not found in database, or search results empty?

i suppose, can't ignore defensive constructs completely, guards in nature defend "normal" flow of app?

is there rule of thumb when defend , when crash?

as fred hebert says @ http://ferd.ca/the-zen-of-erlang.html -

if know how handle error, fine, can specific error. otherwise, let crash!

i'd authentication errors, empty search results, etc., expected errors , warrant appropriate response user.