Backup gyros - which do you trust?

Julian Scarfe wrote:

I think you're barking up the wrong tree here. Classic teaching of
partial/limited panel involves covering an instrument and then continuing to
fly without it. In the case of the Bandeirante accident, that wasn't the
issue. There was still a perfectly serviceable AI in the panel, and a pilot
sitting in front of it. The issue was identifying the failed instrument in a
complex cockpit environment.

I'm not sure I'm barking up the wrong tree.

Possibly practicing flying partial panel makes little sense. OTOH,
practicing partial panel *does* teach which combinations of instruments
can be used to provide the same information as the missing AI.

Surely this is relevant to obtaining and maintaining a good
crosscheck -- and wouldn't good crosscheck be the key to identifying
the failed instrument in a "complex cockpit environment"?

BTW, my reading of the accident report is that they weren't
certain but what both AIs had failed -- something that was certainly
within statistical likelihood given the low MTBUR

Cheers,
Sydney