See if adding a pull-down to RS485 busses prevents powered Mega 5V rail from backdriving through the RS485 into MCM, which then turns fan on.

Need to add pulldown to METSCI_DE... resistor value can be quite high... touching with finger turns it off.  Can't even scope probe signal without changing state.

Didn't need to pulldown BATTSCI_DE... but probably should anyway on next revision.

...

So if we program both LTC1487s to be in "low power mode":
 //METSCI_DE 
 pinMode(2,OUTPUT);
 digitalWrite(2,LOW);

 //BATTSCI_DE
 pinMode(A10,OUTPUT);
 digitalWrite(A10,LOW);

 //METSCI_REn
 pinMode(3,OUTPUT);
 digitalWrite(3,HIGH);

 //BATTSCI_REn
 pinMode(4,OUTPUT);
 digitalWrite(4,HIGH);

...then the relay won't turn on, but we still get voltage leaking into BATTSCI/METSCI rails.  The pullup (R43/R45) and pulldown (R44/R46) resistors (all 1k) tied to the BATTSCI/METSCI lines should inject some current (e.g. 5 volts / 1 k = 5 mA per bus).  This doesn't appear to be enough current to turn the MCM on (enough to energize the fans, at least).  However, it does place a constant 15 mW load - consumed inside the MCM? - on the lithium pack whenever LiBCM's 5 volt rail is on AND the key is off.  

Considering LiBCM pulls 825 mW while running full tilt (with key off), that's less than 2% of that power consumption.

One solution would be to disconnect the +5V rail from METSCI2 & BATTSCI2 with a high-side FET... but that adds quite a few parts.  Need to verify this actually removes the above-mentioned drain prior to adding those parts.  Test:
15 mW is ~3.9 mA thru a 1k resistor (R43).  So we should see 1.1 volts across R43 when the BCM IMA harness connector is plugged in.  

Instead, we're actually seeing 2.36 volts across R43, along with 2.33 volts across R44 (from METSCI1 to ground).  This is due to R40, which - along with R44 & R43 - creates an always-on 2k12 series resistance from +5V to GND.  R40 has 281 mV across it.  Note that BATTSCI's identical circuit has identical voltage measurements across identical parts.

That's 11 mW consumed by these three resistors whenever +5V is on... BATTSCI has an identical load, so in total there's a constant 22 mW load from these biasing resistors whenever +5V is on.

...

So it makes sense then that the pull up/down resistors inject some current (at 2.5 volts no load) into the MCM.

However, we can fix the "fans turn on when Mega unprogrammed" issue by adding a weak pulldown to METSCI_DE.  I've added a 71.5k pulldown to METSCI_DE on the beta boards I've shipped (#004/005/007).

For RevC, need to add pulldown to METSCI_DE & BATTSCI_DE.  

/////////////////////////////////////////////////////////////////////////////////

Solution:
Add high impedance pulldown to METSCI_DE & BATTSCI_DE (e.g. 71.5k).
Add high-side disable FETs between +5V and R43/R45.  Can reuse existing I_SENSOR_EN signal to control.