I can try checking that if I can sandbox the TCU I have - I'm going back tomorrow morning for another one, and at our local junkyard's all-you-can-carry I intend to get at least four 97+ TCUs and 3-4 pre-97 TCUs to play around with along with some ECUs. If you read the SI-5154 datasheet from Sanken you might be able to figure this out just as well as I can (you know way more about transmissions than I do...) - you will want to look at the wiring for pins 2 (input drive signal) 3 (output signal) and 4 (DIAG signal). It looks like I'm not going to be able to figure out what the triple schmitt trigger inverter is for (I suspect it is linked to the output signals somehow) until I try to shave the board down, so that'll have to wait till I can get another TCU tomorrow.
Another thing I really want to do is figure out if the non-populated positions in the connector can be used for something or if they're an engineering debug port or something. Some of them appear to have traces on layers 2 and 3 going to them, but I can't quite make them out because my light table sucks and I haven't even tried to shave the board yet. What I'd really like to do is figure out if the I2C port on the microcontroller that's used to talk to the 24C02 can possibly be expanded to include some sort of display on the dashboard and/or a remote input setup that can override what the TCU would normally do - if I can get a programming adapter for the microcontroller I'm going to see how much it'd take to simply reimplement everything in it, and add cool features with any remaining space.
EDIT: to actually answer your question (oops) it looks like in normal operation the DIAG output of the SI-5154 exactly mirrors the input signal. When the load is open the diag output is high no matter what the input is, and when the load is shorted and/or the chip is overheating, the diag output is low no matter what the input is. When in normal operation the drive output also mirrors the input signal. So it seems to me that the easy way to fake the TCU into thinking all is well is to cut pin 4 on each of the SI-5154s and wire pins 2 and 4 of each SI-5154 together on the bottom of the board, and possibly fake something with the readback of the output pin as well if that's monitored.
EDIT2: it looks like my conjecture about the drivers simply monitoring existence of a load is correct (they only have short/open detect, they can't actually tell how much current is being drawn.) However, looking at the spec sheet they are rated for only slightly more current than the solenoids draw, so your point about using a make-before-break switch is also entirely correct - doubling the current, even momentarily, would risk blowing up the driver transistor. I wonder what magic is contained in the montana fab board that gets around this issue...
EDIT3: further update to parts list - the diode clamps used on the CCD bus lines are NXP PZM15NB2 (11-15 volt zeners) with a top package branding of 152 on a SOT346 package. They are probably there just to keep the NXP 04833637 CCD bus transceiver from being zapped by accidental overvoltage on the bus.
http://tkb-4u.com/codes-and-markings/smd-codes-markings/database was VERY useful in determining this.