Would you in general recommend that policy makers be discouraged from looking at known-to-be-more-accurate sources of any sort? If not, what distinguishes this source?

You haven't said why looking at market estimates would induce more overconfidence when market estimates are under-confident and more accurate than what people would otherwise use.

The longshot bias is an *underconfidence* bias. And since there is almost always "dispersed information that can be aggregated", there is little need to "know" if this happens in any one case.

Your points (i),(ii),(iii), are trivial, obvious, and well-known, and I see no indication that Wolfers and co-authors were trying to make them. They also don't even suggest that looking to betting market prices will induce overconfidence on average. How does the mere possibility of mistakes suggest we shouldn't try to estimate "elusive reality"?