Try both a new hard drive and hard drive controller. One might be causing file corruption.

The controller is onboard, unfortunately, and there's no way to disable it so I can't use an I/O card.

I just booted it up now out of morbid curiosity... Something either in config.sys or autoexec.bat executes an illegal instruction on boot, and forces the system to reboot. I don't understand why it's corrupting itself, but only overnight - I can reboot it 50 times in the day and it'll be fine, but leave it off for 12 hours or so overnight and it seems to be guaranteed corruption on startup. Could it have something to do with a bad CMOS battery?

The CMOS battery only runs the RTC and sometimes keeps BIOS info in volatile memory. It should not be affecting the drives.

Try to run the system for a few days and see if it corrupts itself.

Ampera wrote:

The CMOS battery only runs the RTC and sometimes keeps BIOS info in volatile memory. It should not be affecting the drives.

Try to run the system for a few days and see if it corrupts itself.

I can't do that at this point - one of the binaries being loaded on startup is corrupt, so it's just about safe to say that the system is completely broken again... Should I reinstall DOS and then try?

To be honest, I've got no idea what's going on.
Perhaps it is related to a cold soldering joint or a bad condenser, as both are affected by temperature.
That would at least be an explanation for a system running fine after warm up..

Tip: If you can't find the error, try an older ISA-IDE card instead of the on-board controller (disable that).
A sound card with jumpers and an IDE port can also be used; S7 systems are usually modern enough to boot from secondary master.
If this still doesn't work, run windows via win /b. This should create a bootlog file in Windows directory.

Edit: Also try to manually set PIO modes in BIOS. Can't verify this right now, but I once heard of PIO 4 beeing buggy.

Any bulging caps visible?

It can be caused by a lot of things, most likely a failing HDD, but i have recently been fighting with a motherboard that worked just fine until it booted into Windows with graphics mode other that 640x480 x 16 colours. It would constantly reboot when set to a different mode, no matter what VGA card i would put in there (tried S3 Savage 4, ATi Rage 128, rivaTNT, riva128, all PCI as it was a 486 PCI board). In the end i have decided to replace all electrolytic capacitors on the board. Was a pain but it looks like i was thinking right: Board is booting ok into high res mode, but hangs a minute after. The capacitors i had at hand were not too good, so i will have to redo this job again, but it did help, it seems.
So might be caps.

P.S. all caps looked just fine before replacing.

Best guesses in order of probability:

1) Dying disk
2) Dying disk controller. (Try a SCSI card + SCSI disk, if you can't disable the onboard IDE. Also, some PCI IDE controllers work anyway, even with the onboard still active)
3) Witchety IDE cable
4) Bad caps
5) PSU delivering unstable power

Those last two have odd symptoms, and should be suspected if you have other bizarre behavior - for example, can install Win98 today, but tomorrow the install dies, then randomly starts working again.

Definitely get some sort of PCI IDE controller. They should work great in tandem with the on board I/O.

For example, my Windows 98 box (the full house) has a Promise Ultra66 PCI IDE controller for the hard drives, and I use the motherboards on board io for optical drives.

I'll try giving an external controller in combination with a new disk a go. But first, I'm going to replace the CPU, because I just got another that is not only faster but also known-good as opposed to the current one that came from the bottom of a random drawer at my roommate's parents' house 😜

I tried my first plan last night and eventually decided to replace the CPU with a WinChip C6 until I can get a new Pentium 133. It seems to be working just fine, and I performed a fresh installation of DOS 6.22 and Windows 3.1 and left it overnight. It seems to still be working this morning with no corruption - I'll update this thread if something happens or if it goes a while without anything happening.