Sydney Hoeltzli wrote
What do you think of the conclusions? They seem to be:
1) prevent AIs from failing

Well, there's a lot to be said for that. For one thing, it's far from
unlikely that BOTH of the AI's failed, not just one.

Did you miss this: (All quotes from the referenced report)

"These 21 artificial horizons had an MTBUR of 257 hours." That's mean
time before unscheduled replacement, but... "The artificial horizon
fitted to the EMB-110had no specified overhaul life and was treated as
an 'on condition'item" and thus all replacements were unscheduled.

Why was such a shockingly high failure rate tolerated? Well, "The
BCAR Section under which the aircraftwas certificated did not
stipulate the reliability requirementsthat the artificial horizon
should meet in order to ensure thatthe occurrence of a double failure
was a statistically remoteevent." Gotta love the way those regs
protect us...

And sure enough it was not statistically remote - it had happened
before! "An EMB-110 operated by another UK company suffered two
double artificial horizon failures in 1995. The first,on 4 June 1995,
involved a double instrument failure" There were only a handful of
EMB-110's in the UK...

2) since 2 AIs weren't enough to keep the plane upright (combined
with 2 turn and banks, 2 of every other instrument), require
passenger planes to have 3

Yes, that's the recommendation. IMO it's unmitigated crap. First
off, AI's should not be failing at an average of less than 300 hours.
Second, there were still two good PNI's (basically HSI's) and
turn&slip indicators. But could the pilots use them? Probably not
because "This technique, commonly referred to as 'limited panel' (see
paragraph 1.5.3.2) does not form part of a professional pilot's
recurrency training and testing."

So the most likely causes of the crash are AI failure (quite possibly
double AI failure), and the inability of the flight crew to fly
partial panel because SURPRISE they get no recurrent training in
partial panel flying. Exactly what kind of outcome could one expect
when you fit proven failure-prone AI's to an airplane and don't give
the flight crew any recurrent partial panel training?

(agree, chilling)

There are plenty of chilling accidents out there. This isn't one of
them. This was inevitable.

Michael