Mutual recommendations carry zero information/trust value about the things being recommended. Either it’s a mercenary exchange or they’re real friends or it’s a genuine pair of unilateral recos between strangers (rare enough to be ignored)

actually in the first two cases, if you have independent but different prior trust levels with the two, it actually erodes trust (“I thought A was better than B but if A likes B he either has bad taste in friends or a more mercenary than I thought”)

The third condition could conceivably be detected by socisl network analysis. But then trust discovery should really be attributed to the network analysis.