Interpreting experimental evidence in the presence of post-randomization events: A re-assessment of the Self-Sufficiency Project