possible socket issue?

Give the EPROMs a deeper program cycle.

Don't erase them, perform a few more writes of the same data to get the cells more deeply programmed - and/or increase the program pulse time in your software. A proper write of 2716 chips should take around 90 seconds.

Thanks for the update,
Out of interest, which EPROM programmer are you using?

I always do mine at slow speed, at least twice.

I don't have one of those particular programmers but I believe they default to fastest program speed "+2" which doesn't program some 2716/2732 EPROMs deeply enough.
Someone recently mentioned to me that programmers software also has a "double write" option tick box which should help besides setting the program speed to "-2".

It does.

I have an older GQ-4X (not 4x4) and I never had these issues. I am thinking something changed in the 4x4 version.

I did confirm that the Vpp actually does not touch 25v, but I have not had much trouble and burned probably over a thousand of M2732 ROMs in the last 10 years with that thing. Use the fast speed, double write is on by default. Burns both passes in only like 5-10 seconds. Never had to do multiple burns. Only ones I can remember failing after burn or being unreliable was TMS2732, and I think i just had a bad batch.

Currently go back and forth with a TOP3000 and GQ-4X. Those two burners cover most things up to DIP42. I use the TOP3000 when possible, but it won't do the 24pin ROMs and has a few bad profiles like The 4Mbit EPROM profiles are all screwed up.

I've probably helped half a dozen pinsiders now with weakly written EPROMs that work/don't work depending on where within +/- 10% the 5 volt rail is sitting. That was the giveaway in this thread (they worked cold but failed warm when 5V had drifted a little).
gdonovan was one of the pinsiders I helped and from memory his problems were 2732's. If he remembers, maybe he can comment.
I've had the same issue myself with the TL866A programmer (for the same reason) but more so with 2716 than 2732.

Yeah then something has changed in the new 4x4 because I have not seen that behavior. I burn so many chips I would have ran into that. Maybe should stop recommending that burner? Someone should see whats going on the Vpp pin. I'd guess not going high enough V?

I am not sure how well I'd trust a chip that I had to burn a bunch of times to get the write to stick.

There's a few factors involved like how good the Vpp voltage is, the programming algorithm and chip type. Specification for 2716 generation chips say 50ms programming time per cell which equates to 92 seconds for a fully programmed 2716. So it's no surprise that "fast" 7 seconds full device write times can be flaky and so more write cycles are needed to get deeper programmed cells.

1000s of chips on fast no issue with a 10 year old gq4x

speed may be a factor but doesn't seem to be an issue with mine. My TOP3000 programs most chips even faster than the GQ-4X does. W27C512 is 30 seconds GQ4X and 10 seconds TOP3000.

Any alternatives to GQ4x4 that does 24pin 25vpp EPROMs that is win10 compat and not super expensive?

None that I know of. But at least you can adjust the program time although it's guess work as "-2" through "+2" speed settings say absolutely nothing about how long it will take. Other programmers with speed adjustable settings state the actual program pulse times in micro/milli seconds.
Correct me if I'm wrong but the GQ4X doesn't have the option to verify at +/- 5% and +/-10% of VCC. If it did the EPROMs used here would have failed verification in those steps showing the weak write.
Other thing is we don't know if squad8 friend that programmed the chips used an external power source for the GQ4X so the current from Vpp may not have been high enough to program the cells.

In GQ-4X No VCC control. Speed adjustments are ambiguous, I just leave it wherever the profile is preset too. One thing I have noticed is some chips take longer to program, and if it takes really long I probably selected the wrong profile with too low Vpp. It tells you to use an external power supply, so I always have.

TOP3000 software does not even have a speed adjustment or vcc adjustment. I think it will do a 2716B / 2732B (haven't tried) but no other ones. It has a 25v and 21v profiles, but when you select them the vpp pin stays flat 0v. If it could do the 24pin chips and had a good profile for 27c4001 I would probably retire the GQ4X.

It must be using an intelligent algorithm that program/verifies in short pulses until verify passes. Then it programs an extra pulse for safety. Intelligent algorithms didn't start until 2764/27128 to largely reduce program time. So you're seeing some chips take longer to verify good. Doesn't mean they are bad chips.

Does it expect you to use some sort of socket adapter that you plug a 25V source into?

The TL886II replacement which is the XGecu T56 supports 25V Vpp and has a 48 pin socket. The software lists 2716/2732/27C4001. You can adjust the verify voltage.

The TL866II is cheap enough that it's worth trying to modify the internal buck converter to support 25Vpp. One day...

Lot of projects since that issue, too much water under the bridge and can't recall the details now.