Statistics / social science folks:

When authors of an observational study claim they 'controlled for' various confounding factors, how much should I trust that they actually successfully did?

Never? Only sometimes? (If so, when? Do you have heuristics you use?)