The results from running build/unittest/json_lib/json_normalize-t could be more clear.

For example:

ok 73 - normalize err?

ok 74 - expected '{"foo":"value"}' from '{

  "foo": "value"

}

' but was '{"foo":"value"}'

There are two problems:

1) the TAP context strings are worded such that they only make sense in the failure case, not the (far more likely) success case

2) the input, expected, and actual results are not aligned to make it visiable at a glance if they match or not.