Posted for <edechazal@comcast.net>:

 I did a test yesterday that shed some light on the diving problem.  I was
able to duplicate it in "laboratory conditions".
 
 I spliced a set of jumper leads between the altitude transducer and the
harness wiring.  That means I disconnected the connector at the transducer and
hooked up a 3 wires to the harness connector, looped them out to the cockpit,
then connected their opposite ends to the transducer contacts.  Effectively
just lengthening the harness.  Each looped wire actually had a separable set
of mated terminals at about mid-length so that I could disconnect them and
"fail" the transducer and monitor what happened.
 
 The transducer voltages on the ground (1,000 feet MSL) are 10V power and
approx. 8 V signal.  At 2,500 feet the signal voltage is 7.6V.  S-TEC told me
awhile back that 5V is something like 13,000 feet.  Flying at 2,500 in light
turbulence, everything performed properly with altitude hold engaged.  The
signal on the digital multimeter appeared fairly steady although the last
readout digit (hundredths of a volt) did fluctuate a bit.
 
 Things got exciting when I broke the signal circuit.  I broke it and nearly
immediately the servo pushed the stick forward.  Also, the voltage on the
"harness" side of the break went to 4.9V.  After 2 seconds of this, I canceled
the autopilot and returned to level flight manually.
 
 Next, to simulate an intermittency, I broke the circuit for 1 second (the
pitch forward occurred again) then I reconnected the line.  The autopilot
nearly immediately went from pitch down command to very aggressive pitch UP.
 I allowed this only long enough to see that the autopilot was not going to
intercept the original altitude, but blow past it with no apparent reduction
in the pitch up command.
 
 My passenger at this point was beginning to turn green around the gills so
testing ceased.
 
 A couple of notes:  The disconnected voltage on the harness side of the
signal circuit is supposed to be 5V according to S-TEC.  My 4.9V was close
enough, I judge.
 
 So how to explain all this?  One hypothesis is this:  In this test the
computer was happy seeing 7.6 V out of the transducer.  When I broke the
circuit, the computer saw 4.9V all of a sudden (at least that's what's on the
multimeter).  It thinks "hey, how did I all of a sudden get to 13,000 feet
when I was supposed to be at 2,500 feet? I better pitch down in a hurry."
 Makes sense to my "human" computer, but I have no idea if the STEC computer
would see it the same way.
 
 This suggests the cause of the dive/climb problem some of us are having is an
intermittent break in the signal transducer circuit.  I should note that I'm
getting better performance (before it was like what Scott describes - a few
seconds of control, then departure) when I cut off the standard harness
connector (white 3-pole Molex style) and crimped higher quality automotive
contacts onto the wires, shrink tubed them and connected each directly to the
proper male pin on the transducer.  The original contacts are gold, but the
shape of the female contact allows the barrel to open up over time and a bad
connection could arise.  But even after this change, I have had one instance
of a Dive command.  Also, I have not tested what happens if the break is in
the ground or power circuit.
 
 Anyone have ideas to throw on top of all this?
 
 Regards,
 Ed de Chazal