But the entire corpus for Linear A is tiny, you could backfill a "translation" that has no actual similarity with the real language when properly tested agains

But the entire corpus for Linear A is tiny, you could backfill a "translation" that has no actual similarity with the real language when properly tested against novel examples. How was this tested?