Contract-focused tests and boundaries

The idea

A meaningful test states a requirement and checks its observable consequence. For a stable filter, test which values remain, their order, their duplicate counts, and whether the original input changes. These are separate promises. A correct length alone cannot establish a correct result, and a correct result alone cannot establish that the caller’s data was preserved.

Boundary cases target common mistaken implementations. An all-negative maximum exposes initialization to zero, consecutive deletions expose shifting indices, and equal best candidates expose tie handling. Include an empty case when it is supported and an explicit rejection case when it is not. If a method reports invalid input with a particular exception class, check that class rather than accepting any unrelated failure as a successful test.

def keep_positive(values); values.select { |n| n>0 }; end
input=[2,-1,2,0]
result=keep_positive(input)
raise "wrong result" unless result==[2,2]
raise "mutated input" unless input==[2,-1,2,0]
puts "checks passed"

checks passed

The checks preserve both repeated positive observations and the untouched original input.