Ok, so one of the problems IS downtime related

So, using an alt I setup a high sec test bed and decided to do some robotics production from materials I got out of wormhole space. Here is the starting position:

At start

For these purposes we can ignore the extractors feeding the storage that feeds the basic processor that feeds the starport. The important part is that there is raw material in the NE storage (YT-UXT). These are routed to 4 advanced industrial facilities. Two are making Mechanical Parts. The other two are making Consumer Electronics. All the outputs are routed back to the storage. These are in turn routed to the fifth Advanced Industrial Facility to make Robotics. These are then directly routed to the starport. There are enough materials for 20 full cycles of the entire system. All 20 cycles worth of material start off IN the storage.

After Starting it up

This is after everything is kicked off. Note that there are enough materials for 18 cycles in the storage. The 4 advanced industrial facilities producing the P2 materials have a cycle in production and a cycle in the input hoppers. So far all cycles are at least accounted for.

Before Downtime

This is the situation just before downtime. 11 cycles of material in the storage. 1 in the 4 P2 processor's production, 1 in the 4 P2's input hoppers, 1 in the P3 production cycle and 6 cycles worth of final P3 output in the starport. Again at this point all materials are accounted for.

A half hour after downtime

Now this is a different situation. It's a half hour after downtime. This is about 3 h and definitely 3 cycles later. Note that we have 8 cycles of materials in the starport as expected. The Production cycles and input hoppers account once again for a total of 3 cycles of material. However there are only 7 cycles at the startport! There should have been 9!... This means that two entire cycles worth of materials have been blown away entirely.

After all cycles have finished

Sure enough there we go at the end of the production run we are missing two entire cycles of output! Thanks CCP!

My guess is that if an output cycle happens exactly when TQ is down the ouput is "lost", but the input is consumed correctly. TQ these days seems to be down for only about 1/2h And you need to be exact about when the cycle hits. This may be why when I started a cycle 5min before downtime it was fine and with no loss. When the next cycle it that time TQ was up. This time I seem to have nailed it.