Maybe rethink the gun. They have these things that use rapidly (but, crucially, not instantly) expanding liquid nitrogen to propel sensitive payloads real fast. I saw a study about using them to throw bombs on mountains for avalanche prevention. That solves the temperature problem and maybe the brittleness problem.

Also, there is an ice-adjacent material that could make the bullet named pykrete which is more durable than pure ice (Mythbusters made a boat made out of it one time). That defeats the purpose of an ice bullet somewhat, but who knows, the added sawdust might just confuse the forensics.

Though if you really wanna kill someone with an ice projectile, a big hunk of ice thrown from a sling is more than enough. Bonus points if you give it two sharp edges like the romans did.

I spent way too long thinking about this.

An ooblek bullet.

I wonder if it's something that used to work when guns were less consistently mass-produced

Or maybe it's always been nonsense, never knew that though thanks

I feel like the article you linked doesn't support your conclusion, at least for the technique described in the article as an improvement over what people were doing before 2013. Those NIST researchers seemed to conclude that their 3D scan techniques can reduce the false positive rate to very low numbers, even when comparing 9mm rounds fired from consecutively manufactured handguns of the same model. At least if they recover an undamaged bullet that didn't get mangled by the actual shooting.

But yeah, the previous method sounds about as reliable as the My Cousin Vinny expert testimony: maybe getting things down to a range of possible models, but not specifically identifying a specific gun.

Now I kinda wish I had a mythbusters budget for comparing bullet and casing markings to both replicate the NIST study and to just compare whether different manufacturers have very different markings for the same caliber.