Which PLDs were available in Loki's time ?

[Nasta]

Xilinx had just come out with the first XC2000 and 3000 chips around that time, I think.
Altera EPLDs of that time were FAR too simple to do anything really ULA like, but you could do some quite serious (if not very fast) stuff with Xilinx. Xilinx was also VERY expensive.
Miracle Systems used Altera EP1810 for the GoldCard, SuperGoldcard and QXL board. EP1810 is sort of a huge GAL, like 64V48 and is one of the very rare fully interconnected CPLDs, nowadays it's all logic blocks which makes the job of the fitter more complex as well as calculating delays and assigning pins but at least in theory crams in more fast logic.
Altera produced a die-shrink version of the EP1810 later on and Miracle had all sorts of problems getting them to work with their older config files because of subtle timing differences (they used combinatorial latches to switch some features using the address bus, to reduce logic usage).
If you find EP1810 on the market it will be the OTP version. The bigger problem is getting the development tools. Finding that would be of some use to the community because some partial sources of the GC and SGC Ingot (EP1810) exist and some details of the function are not exactly known, though one could say it's not worth bothering about, rather re-construct the logic from scratch (where the aforementioned details of function may prove useful, though).
OTP chips WILL get erased by X-rays (given the right X) but the functionality of the chip after that will be questionable. I suppose with enough chips to try it out, one could get to the right erasing dose that is not lethal to the chip. The problem is that too much knock off ions from neighbouring material (including metallization) inside the chip where they contaminate stuff and you get increased leakage.

[Nasta]

It would be perfectly logical to get the logic done in PLDs at first, but it would never have been commercially viable doing the production model that way. A semicustom chip would have been made, by that time standard cell semicustom manufacturing was already widely used and well within mass market prices for large volumes,

[Nasta]

It would be perfectly logical to get the logic done in PLDs at first, but it would never have been commercially viable doing the production model that way. A semicustom chip would have been made, by that time standard cell semicustom manufacturing was already widely used and well within mass market prices for large volumes,

So by your estimate what should be roughly fair eqivalent of what could be done as smeicustom ?
Were ther options that would offer memory onboard ?
Woudl something within lattice isp4000V line be fair option for QL Parallel TImeline replica with 68000 ?
Or perhaps Lattice's ice40HX ? Or Altera's MAX5 or 10 ? Or something else, (preferrably with long-term support ) ?

If I understand this correctly, you are asking what the modern day equivalent to something like a standard cell chip of that time would be?
Definitely a FPGA rather than CPLD, especially since there were some on-chip memory options. Difficult to say what family, but it seems to me Xilinx would be the more suited one, even the older families. Now, continual availability... that's always a problem. Although some advanced standard cell tech offered an ability to integrate whole MCUs, RAM, ROM, standard peripheral chips etc, I have no doubt these would be WAY too expensive for sinclair. So it would probably be the standard library cell designs *based on standard logic family devices plus some more primitive and more advanced stuff like FF arrays or generalized gates and registers.