As an update, for anyone insterested, I haven't resolved the problem but have discovered that it only affects video RAM that relates to two of the four chips. The corrupted text contains a pattern, with even bits more likely than odd bits to be randomly on. It behaves almost as though there should be a pull-down resistor on the data lines between VRAM and character generator that are absent and resulting in a floating output, but the schematic doesn't reflect that there are any such resistors. Verified the ground connections on all those RAM sockets and they're all fine.
I verified that the slightly out-of-spec crystal is not the cause of the problem, by hotwiring the oscillator from the working Kaypro II into the work-in-progress Kaypro and observing no change.
The +5V out on the power supply is only giving +4.4V and the datasheet for those memory chips states a minimum of 4.75V, but I wired it into a 500W ATX power supply temporarily and the problem did not go away when it was given correct voltage.
I started replacing the ceramic capacitors but so far the only ones that I pulled that were out of spec were unrelated to that part of the board.

I have 2 of these systems that I have yet to tear into. I wish I could be of more help, but this forum is dedicated mostly to gaming on 486/pentium systems.

I would recommend contacting the greybeards over on: https://www.vcfed.org/forum/forum By their standards, Kaypro 2's are rather "new". ^.^

I don't know the KayproII, but think I found a schematic for it. It looks to me like the line select is handled by the LC[0..4] signals, allowing for up to 32 lines. LC[0..3] end up being address inputs in to what I think is the video ram, but LC4 looks like it gets used as a bank select between (U29 & U30) and (U28 & U31). So maybe the bank select toggle U15 (NAND gate wired as NOT, pins 9,10 & 8) has failed, so the second bank of video RAM is never enabled, so the address select to the character set PROM is floating for all lines past 16.

LC4 should come from U11 pin 6 (top right, sheet 2) to U18 pin 3, then from U18 pin 4 to U16 pin 2, then out from U16 pin 4 to the enable pins (pin 8) on U29&30 and the inputs to the U15 NAND gate.

luckybob wrote on 2021-08-06, 08:13:

I have 2 of these systems that I have yet to tear into. I wish I could be of more help, but this forum is dedicated mostly to gaming on 486/pentium systems.

I would recommend contacting the greybeards over on: https://www.vcfed.org/forum/forum By their standards, Kaypro 2's are rather "new". ^.^

I actually just created an account there yesterday so I could see the pics from someone's Kaypro II repair chronicles. It wasn't as helpful as I had hoped but I'll post there also and see if anyone has any ideas.
Thanks

snufkin wrote on 2021-08-06, 12:29:

LC4 should come from U11 pin 6 (top right, sheet 2) to U18 pin 3, then from U18 pin 4 to U16 pin 2, then out from U16 pin 4 to the enable pins (pin 8) on U29&30 and the inputs to the U15 NAND gate.

I had found the same thing on the schematic and scoped that pin out last night and verified that it's at least toggling both of the two banks. I swapped U15 and U18 with known-good ICs from another Kaypro but it had no effect. It appears to me that it is reading both banks (my thoughts about a floating output were bunk because it'd affect both banks if that were the case) but for some reason, not able to write to U28/31.

I think what I'll do is I'll fish out a DIP-8 socket from my box, and I'll mangle it in various ways and put it between the IC and the sockets to interfere with each address line and the bank select line one by one on my good Kaypro to see if I can reproduce the problem there, and thereby identify exactly which address line is the issue, if it is an address line

I found the issue. An amputee, U60, was missing pin 7. I changed it out when the screen was entirely dead, before I discovered the clock crystal was dead. So there's a reasonable chance that I didn't align the chip to socket correcly when swapping and did not notice I broke the pin.

Rather than hiding my shame by pretending I didn't just spend many many hours troubleshooting a broken pin on an IC, I'll leave it here for Google to index for the next sorry soul who finds a similar issue.