I certainly am not proposing putting in a field upgradable design just to enable the ability to deliver poor quality firmware. And I can certainly understand the challenge in adding such capability. My point is just that if it is feasible, it would be a nice feature to add.
I also agree with the 34401A stability, I have two with the final firmware, and those meters are rock solid, but that came with a decade + of updates. Mine has the final firmware, version 11, I am not sure how much was fixed in the previous 10 versions, but clearly even in the case of 34401A, HP/Agilent saw a need for firmware upgrades over the years. However, in the case of the 3446xA, they were able to add capacitance measurement to the early build later on, and each firmware release addressed uncommon problems that I will never encounter in my use. It is nice to know that I have patched it so it will not happen to me if that use case were to come up.
Anyway, back to the 786, I was able to get some random numbers to come up by shorting the leads, but with 50+ attempts on those ranges, I was never able to get an erroneous reading that last for more than a second. They were all transient readings that lasted for less than 1 second.
I can actually make the same behavior occur on my Fluke 179 with the latest firmware in manual range. With the two 34401A I own with firmware 11-05-03, they also generate a split second random reading in manual range. The only difference is these meters goes back to 0 faster, where 786 could sometimes takes close to 1 second.
The only meter that shows the right reading in manual range are the 34461A and 34465A, they both display the proper value at every try.