there's plenty available on yt.

Here's a nice channel with some (close to 400)
https://www.youtube.com/@arnlol

Thanks for sharing - unfortunately Arnlol's video aren't ideal, as they don't really have any long periods of idle

Ideally don't really just want to nab someone else's work off YouTube, which is why I'm actively appealing for people who are happy to share 😀

I'm not saying you should just take the sound from his videos. You just ask him.
The problem with the idle sound you mention is that it changes with the age of the drive, the wear on the spindle bearings.
A new drive sounds very silent while a worn drive tends to be an annoyingly pitched whine.

Looking at the project, you probably want more than just one sound type for access.

Looping a single .wav file is not really going to be realistic.

Synchronous access sound is going to be a lot different that random access sound. You need the sound of the head(s) moving back and forth with random access and the solid grrr noise for synchronous access.

I think creating disk access (read) pattern will simplify audio recording/editing.
It must be a DOS/Windows program that will access sectors in a certain way.

For example:
- The disk (heads) makes a single access + short pause + disk sequentially reads several sectors + long pause
- Moving to the next N sectors
- Repeat

This way you will get dozens of WAV-samples to simulate real random/sequential reading.
Audio cuts with long pauses can be used to randomize idle sounds.

To model the sounds of an HDD, I'd expect there to be a more-or-less continuous spindle sound, and then on top of that would be sounds from head activity, depending mostly on the distance of a seek, and possibly the head positions. Very old drives had stepper motors instead of servos, and had the same number of sectors on every track. Most IDE drives have more sectors on the outer tracks, so they'd tend to spend more time on one of those tracks before stepping to the next one.

Anyway, here is a nice WD Caviar 2635 booting up, and doing a SNOOPER disk benchmark at the end. http://www.hyakushiki.net/junk/wd2635.wav

maybe use a contact microphone for recordings ?

Or create a real synth algorithm, who generates it on the fly ?