Most indicators are useful in one direction.
For example, if my blood pressure is really high, that's a good indicator I have a health problem. But the converse is not true -- normal blood pressure does not prove good health.
Unemployment statistics are underinclusive, so a high unemployment statistic is clearly a problem. In contrast, a low unemployment statistic does not prove that there isn't a problem; the stat isn't counting all those out of work.
CBO estimates are based on the bill sponsor's own assumptions, so if even the CBO agrees a bill will be expensive, it will be. In contrast, if the CBO says that, based on the assumptions given it, the bill will not be expensive, then the CBO's estimate is likely as accurate as its assumptions.