How can an execute_script line work when run in a debugger but not when in the script itself?

So I have the following (mildly-anonymised) evaluate_script line in a javascript capybara test:

page.execute_script("$('selector').addClass('-class')")

It isn't working - e.g. something like page.execute_script("console.log($('selector').hasClass('-class')"), prints false.

if I put a debugger (binding.pry) after that line, and then run the addClass line in the debugger -- literally the same line that's just executed; copied and pasted -- then it does work.

It doesn't seem to be a timing issue -- putting sleeps after the addClass line doesn't change anything.

It's very repeatable: the exact line that doesn't do anything in the script then suddenly works if I replace it with a binding.pry and run the line from pry. Which AFAIK should be impossible.

So: what could cause the same thing to work when run from a debugger, and not when run not from the debugger?

(Capybara driver is poltergeist on phantomJS)

Cheers :)